Claude (Anthropic) for Automation

Key takeaways

All facts are as of 12 October 2026; prices, retention, residency and benchmark facts were re-checked against primary sources on 12 October 2026. Benchmark scores are vendor-reported unless marked independent. INR figures use ₹96 per USD, the X-Rates average for October 2026.

• Four current API models on one price ladder. Per million input / output tokens: Haiku 5.5 $0.10 / $0.50 (₹9.6 / ₹48), Sonnet 5.5 $2 / $10, Opus 5.5 $4 / $20 and Fable 5.1 $10 / $50. All have a 1M-token context window and 128K-token output limit. Anthropic tells most teams to start with Opus 5.5. Mythos 5.1, priced like Fable 5.1, is for vetted organisations only.
• Seven launches in four months (9 June to 7 October 2026), several with breaking API changes. Anthropic commits to no retirement within 12 months of launch for each current model. Budget a migration for each system built on Claude.
• Discounts decide the bill. Batch halves prices, cached input costs 2.5% to 10% of base, and both stack with the 1.1x US-only surcharge. Watch Haiku 5.5's fivefold price step above 100K prompt tokens and a tokenizer that yields about 30% more tokens on Claude 4.7 and later.
• The automation stack is mostly out of beta. Computer use and Agent Skills went GA on 19 August 2026, the day a browser toolset launched. The Agent SDK's code is MIT-licensed but its use falls under Anthropic's Commercial Terms, and the proprietary Claude Code had 56.3M npm downloads in the month to 10 October. Managed Agents ($0.08 per session-hour) is still beta and excluded from zero data retention and HIPAA BAAs.
• Anthropic's own API offers only "global" or "US" residency, and only the previous Claude generation can be kept in India on a cloud. AWS Bedrock's India geography profiles (in. prefix; Mumbai and Hyderabad only) cover Opus 5, Sonnet 5 and Haiku 4.5. Haiku 5.5, Sonnet 5.5, Opus 5.5, Fable 5.1 and Mythos 5.1 are global-only from the India regions, and no Claude model has India processing on Google Cloud.
• Opus 5.5 leads most rows of Anthropic's own benchmark tables and tops Epoch AI's independent index (167, 1 October 2026). Anthropic reports no SWE-bench or tau-bench result for any Claude 5.x model, and none was found elsewhere.
• Scale: run-rate revenue rose from $14B in February 2026 to over $47B in May (both Anthropic figures) and a press-reported $65B in July. The May Series H raised $65B at a $965B valuation, and a confidential draft S-1 followed on 1 June.
• India is the second-largest consumer market (5.8% of Claude.ai use, Anthropic India brief of 16 Feb 2026; the June 2026 data release gives India 6.92%–7.12% of use and a per-capita index of about 0.3) but ranks 101st of 116 per capita. A Bengaluru office opened on 16 February 2026. One outlet reports rupee prices for consumer and Team plans, GST included, from 13 July 2026; the API is priced in USD only.
• Model access itself can be cut off. A US export-control directive forced Anthropic to switch off Fable 5 and Mythos 5 for every customer from 12 June to 1 July 2026, and the BBC reports that the US Department of Defense has stopped using Anthropic products. Keep a tested fallback model.

Seven launches in four months produced a four-tier 5.x lineup

Anthropic's platform release notes give the launch dates. Opus 4.8 shipped on 28 May 2026. It was followed by Fable 5 and Mythos 5 on 9 June, Sonnet 5 on 30 June, Opus 5 on 24 July, Fable 5.1 and Mythos 5.1 on 1 September, Opus 5.5 on 22 September, Sonnet 5.5 on 28 September and Haiku 5.5 on 7 October (release notes). The models overview lists four current models. Each has a 1M-token context window, a 128K-token output limit, text and image input, and adaptive thinking (models overview). For the full model chronology across vendors, see the AI automation timeline.

Sources: release notes, models overview, pricing.

Anthropic's advice is to "start with Claude Opus 5.5 for most workloads," to use Fable 5.1 "for demanding reasoning and long-horizon agentic work," and Haiku 5.5 "for high-volume, latency-sensitive tasks such as classification, extraction, and routing" (models overview). Anthropic says Opus 5.5 "performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5." Its list price is only 20% lower, so the rest is a per-task efficiency claim (Opus 5.5). Sonnet 5.5 is said to run 30%+ faster than Sonnet 5 at up to 30% lower cost per task, and is the first Sonnet with cyber safeguards (Sonnet 5.5). Haiku 5.5 "costs about 75% less to run on average than Haiku 4.5," although its list price is 90% lower (Haiku 5.5). Opus 5, launched in July as "close to Fable 5 at half the price," became the default on the Max plan (Opus 5 announcement).

Opus 4.1, Opus 4, Sonnet 4 and Haiku 3.5 are retired except on some clouds; most other 4.x models remain available (pricing). Sonnet 4.5 retires from the Claude API on 30 November 2026 (release notes).

Each launch changes the API. On Haiku 5.5, manual budget_tokens, temperature, top_p, top_k and assistant prefill "can each return a 400 error." On Opus 5.5, thinking cannot be disabled, and forced tool choice (any or tool) returns 400, as it also does on Sonnet 5.5. Older computer-use tool versions return 400 on Haiku 5.5, Sonnet 5.5 and Opus 5.5 on the Claude API and Google Cloud, while Bedrock still takes computer_20251124 (release notes, re-read 11 October 2026). Fable 5.1 accepts only adaptive thinking, errors on forced tool use, and invalidates thinking blocks when earlier turns are edited (Fable 5.1 model page). Code written for Claude 4.x needs changes. Any business running Claude integrations should pin model IDs and run a regression suite at every upgrade.

List prices span 100x, and the discount stack matters more than the sticker price

Input and output prices for the current models differ by a factor of 100 from Haiku 5.5 to Fable 5.1. Cached input differs by only 25x, and the discount rules change the comparison more than the list prices do. All prices below are USD per million tokens from Anthropic's pricing documentation. The INR columns are conversions at ₹96 per USD, because Anthropic publishes no INR API prices.

Sources: pricing (re-read 11 October 2026; fast-mode cache figures apply the documented multipliers to the $8 fast-mode input price), models overview, Opus 5.5, Haiku 5.5. OpenRouter's records match Anthropic's prices on every row it lists (OpenRouter models API).

Four multipliers do most of the work

Message Batches cuts input and output prices by 50%, with up to 100,000 requests per batch and, under a beta header, up to 300K output tokens per request on the Opus, Sonnet and Haiku 5.x models (models overview, rate limits). Prompt caching is the biggest lever for agent loops. Cache writes cost 1.25x base input (five-minute cache) or 2x (one hour). Cache reads cost 0.1x on most models, 0.05x on Opus 5.5 and Sonnet 5.5 (a 95% saving on repeated input) and 0.025x on Fable 5.1. For most models, cache reads also do not count toward the input-tokens-per-minute limit, so caching raises throughput too (pricing, rate limits). US-only inference costs 1.1x, and regional cloud endpoints carry the same 10% premium. The multipliers stack: Sonnet 5.5 through Batch with US-only processing costs $1.10 input and $5.50 output (pricing). Fast mode (research preview) costs $8 / $40 on Opus 5.5 (2x) and $10 / $50 on Opus 5 and 4.8, runs only on the first-party API (not Bedrock or other partners) and cannot be combined with Batch (pricing).

Four overheads catch budgets out. Haiku 5.5's price rises fivefold for the whole request once the prompt exceeds 100K tokens, and cached tokens count toward that length (pricing). The tokenizer on Claude 4.7 and later produces about 30% more tokens for the same text, so cost tests on older models understate spend. Tool definitions count as input: the tool-use system prompt adds 286 to 406 tokens on the 5.x models (up to 804 on Opus 4.7), the computer-use toolset about 4,500 tokens per request and the browser toolset about 6,600, or roughly $0.018 and $0.026 per uncached Opus 5.5 step. Server tools bill separately: web search costs $10 per 1,000 searches plus result tokens, and code execution is free alongside web search or fetch but otherwise billed per container-hour (pricing, claude.com pricing). Thinking tokens are "billed as output tokens, even when the thinking text isn't returned" (extended thinking).

Seat plans cluster at $20, with API credits bundled at the top

Source: claude.com pricing. After the Opus 5.5 launch, Anthropic raised five-hour usage limits on the Pro, Max, Team and seat-based Enterprise plans (Opus 5.5). Anthropic's pricing pages show USD only; the INR column is a conversion at ₹96. One outlet reports rupee prices for Indian users from 13 July 2026, with GST included in displayed prices: Pro ₹2,399 a month billed monthly (₹2,000 a month on annual billing), Max 5x ₹11,999, Max 20x ₹23,999, Team Standard ₹2,399 per seat a month annual or ₹2,999 monthly; UPI is not supported (single-source, TechStory). INR invoicing for the API is unresolved.

What a task costs at list price

Model tokens are a small part of most automation costs. The figures below are token arithmetic from the cost research in this set (AI automation costs), using assumed profiles at standard prices without Batch, retries, evaluation or human review. The cost research suggests multiplying by 1.2 to 2 for a production budget, itself an assumption.

Source: cost arithmetic from AI automation costs at Anthropic list prices. The cost research priced the Haiku coding run at $0.136 without the >100K step; with context growing evenly from 20K to 120K tokens, the last 8 of 40 turns bill at the higher rate, roughly doubling it to about $0.28 (arithmetic in this document). Haiku 5.5 ties OpenAI's GPT-6 Luna, which lists the same $0.10 / $0.01 / $0.50 prices (OpenAI pricing). Intercom charges $0.99 (₹95) per AI-resolved conversation and Zendesk $1.50 to $2.00 per automated resolution, which is 19x to 38x the Sonnet 5.5 model cost of a five-turn resolution (Intercom pricing, Zendesk pricing). Zendesk's price varies by regional page (India and UK: £1.30 committed, US$2.00 pay-as-you-go, 11 October; Germany: €1.50 / €2.00, 12 October); the US $1.50 committed figure is an earlier read plus secondary sources. Tooling, support and integration, not tokens, make up most of the gap between those fees and the model cost. In absolute terms, model choice matters most in long coding loops ($0.28 to $4.96 per run). Elsewhere every tier costs cents or less per task.

Spend caps and per-model pools, not model quality, halt unattended pipelines

Anthropic's rate-limit page has five tiers: Evaluation (a possible starting tier for new organisations, below standard limits), Start, Build, Scale and Custom. The Start, Build and Scale tiers each carry a monthly spend cap; Custom has none, and none is listed for Evaluation. When an organisation hits its cap, every request returns HTTP 429 with no retry-after header until 00:00 UTC on the 1st of the next month, unless it moves to a higher tier sooner (rate limits, re-read 11 October 2026).

Source: rate limits, re-read 11 October 2026. RPM is requests per minute; TPM is tokens per minute.

A $500 Start-tier cap pays for about 9,600 five-turn Sonnet 5.5 resolutions, 4,800 on Opus 5.5, 187,000 on Haiku 5.5, or about 200 forty-turn Sonnet coding runs, so one busy workflow can halt every automation on the account mid-month. Limits apply per model, Fable 5.1 and Fable 5 share one pool (the two Mythos models share another), and limits are shared across residency settings, so switching models changes throughput. Managed Agents adds organisation-wide limits of 300 create and 1,200 read calls per minute (rate limits). Claude Platform on AWS bills through AWS Marketplace at $0.01 per Claude Consumption Unit and has no tier-increase flow, per-workspace limits or fast mode (pricing). Reserved capacity is no way out: Anthropic says capacity commitments are "no longer available for purchase", and Priority Tier is not supported on Haiku 5.5, Sonnet 5.5, Opus 5.5, Opus 5, Sonnet 5, Fable 5.1 or Mythos models; no price is published (service tiers). Outage history could not be checked, because the status page returned no incident rows. Build queueing, budget alarms and per-workflow spend isolation into every pipeline rather than relying on retries.

Data residency stops at "US or global" on the first-party API

The Claude API's one residency control is a per-request inference_geo of "global" (default) or "us". The US option costs 1.1x on every token category and works on Claude 4.6 and later. Workspace geography is US-only and fixed at creation. On Bedrock and Google Cloud the endpoint sets the region, and on Foundry the US option is a "US Data Zone Standard" deployment (data residency). The documentation has no India or Asia-Pacific value for inference_geo. Anthropic's regional compliance page goes further: "Regional data residency is available across Europe, United States, Canada, and Asia-Pacific regions including Japan, South Korea, Singapore, India, and Australia," via regional endpoints on Bedrock, Vertex and Foundry (regional compliance, read 11 Oct 2026; not re-readable 12 Oct, JS-rendered). It names no India region ID, and it lists Sonnet 5, Opus 5 and Haiku 4.5 as the models for regional deployment. AWS's own Bedrock documentation (read 12 October 2026) bears the claim out for exactly those three models: India geography profiles (in.anthropic.claude-opus-5, in.anthropic.claude-sonnet-5, in.anthropic.claude-haiku-4-5-20251001-v1:0) route only to ap-south-1 (Mumbai) and ap-south-2 (Hyderabad). Opus 5 and Sonnet 5 are also listed In-Region in Mumbai on the bedrock-mantle endpoint on AWS's regional-availability page, but the two model cards omit Mumbai (confirm in the console). The current generation (Haiku 5.5, Sonnet 5.5, Opus 5.5, Opus 4.7/4.8, Fable 5/5.1, Mythos 5.1) is global-only from the India regions. AWS adds: "Don't rely on the geographic prefix ... alone" (call GetInferenceProfile), and prompts and outputs "may be stored in the opt-in Regions for abuse detection purposes" (Hyderabad is opt-in, so still India) (AWS regional availability; Sonnet 5 model card). On Google Cloud no Claude model has India processing; its data-residency page lists Singapore ML processing for Opus 5.5, Sonnet 5.5, Sonnet 5, Opus 5, Fable 5.1 and Fable 5, but the locations page does not (unresolved) (Google Cloud data residency). On Foundry only a US Data Zone option is documented for Claude (Azure deployment types). Treat India-hosted inference of the 5.5, Fable and Mythos models as unavailable.

Sources: data residency, regional compliance, Claude in Amazon Bedrock, Claude on Google Cloud, AWS regional availability, Google Cloud data residency, Azure deployment types, API and data retention, Opus 5.5, Anthropic ZDR article, release notes, Managed Agents, Enterprise Frontier Safeguards, Claude memory help.

Managed Agents can be pinned to a geography per agent or session at the 1.1x US rate, and self-hosted sandboxes keep tool execution on the customer's infrastructure. Memory is the exception. Since 19 August 2026, self-hosted sandbox sessions can attach memory stores, but the data-residency documentation says their contents "remain stored by Anthropic and are copied to your sandbox for the session" (data residency, release notes, both re-read 11 October 2026). Agent memory therefore sits with Anthropic even when execution does not. Under the Commercial Terms, the customer owns outputs, Anthropic may not train on customer content, paid customers get an IP indemnity (excluding customer modifications, combinations with non-Anthropic technology and customer inputs), and liability is capped at the previous 12 months' fees (Commercial Terms).

For businesses handling Indian personal data there is a deadline. The DPDP Act's consent-manager provisions start on 13 November 2026. Its core obligations on notice, consent, security, breach notification, data-principal rights and cross-border processing start on 13 May 2027 (G.S.R. 843(E)). Rule 7(2) requires a detailed breach report to the Board "within seventy-two hours", and Rules 6(1)(e) and 8(3) require logs to be kept for one year; cross-border transfers fall under Rule 15's conditions and the Act's s.16 power to restrict transfers to notified countries, all from 13 May 2027 (G.S.R. 846(E)). Before then, data-processing agreements and records should state Claude's processing location, retention mode (zero or 30 days) and whether Anthropic holds agent state. See India and governance.

The automation stack left beta on 19 August 2026, except Managed Agents

Anthropic now sells a full agent stack, from raw tool calls to a hosted agent runtime. Most parts are generally available. The exception is the managed runtime most suited to unattended, long-running jobs.

Sources: pricing, computer use, release notes, programmatic tool calling, tool search, memory tool, Agent Skills, Managed Agents, Agent SDK, Claude Code GitHub Actions, Claude in Chrome, claude.com pricing.

Managed Agents gained session budgets with a hard spend cap and an "advisor model" (7 August), an auto permission policy in which the server evaluates tool calls (10 September) and beta "dynamic workflows," where an agent writes a program that runs many agents in phases (9 October 2026) (release notes, Managed Agents). That leaves two architectures. The Messages API keeps the loop, state and data path under the builder's control and is the only route for ZDR-bound or BAA-bound work. Managed Agents removes the infrastructure work for long-running asynchronous jobs, at the price of server-side state. Anthropic's engineering guidance still says "the most successful implementations use simple, composable patterns rather than complex frameworks," now with a note pointing readers to Managed Agents (Building effective agents). See agent frameworks and tool stack.

On 9 December 2025 Anthropic donated the Model Context Protocol (MCP), the protocol behind Claude's connectors, to the Linux Foundation's Agentic AI Foundation (AAIF), to which Anthropic (MCP), Block (goose) and OpenAI (AGENTS.md) contributed the founding projects. Its Platinum members are AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft and OpenAI (Linux Foundation). Anthropic then reported over 10,000 active public MCP servers and 97M+ monthly SDK downloads (December 2025 figures, vendor-reported) (MCP donation). A full pass of the official MCP Registry API on 12 October 2026 found 42,172 distinct servers (41,656 with an active latest version); these are registry entries, not production use (MCP Registry API). The 28 July 2026 specification made servers stateless, and the Tier 1 SDKs see close to half a billion downloads a month (MCP blog). Enterprise-Managed Authorization, stable since 18 June 2026, lets an identity provider grant MCP access at first login (enterprise-managed auth). Anthropic bought Stainless, "a leader in SDKs and MCP server tooling," in May 2026 (Stainless acquisition). Connector work built on MCP carries over between vendors.

Agent security problems are design constraints, not bugs awaiting a patch. The computer-use documentation warns that "Claude will follow commands found in content even when they conflict with your instructions" and tells integrators to isolate Claude from sensitive data and actions (computer use). Anthropic says Opus 5.5 "has a classifier that screens every action before it runs" (vendor claim; Opus 5.5). Yet Simon Willison's log records Johann Rehberger bypassing Claude Code's default "auto mode" about 80% of the time by getting it to download, unzip and run code, in an entry dated 27 August 2026, and recommends containers, restricted egress and no credentials inside the agent's runtime (Willison prompt-injection tag). Earlier research showed agents being hijacked through a malicious GitHub issue (Invariant Labs) and steered by poisoned MCP tool descriptions (Invariant Labs).

Claude Code and the Agent SDK lead on installs, not GitHub stars

Claude Code shipped with Claude 3.7 Sonnet on 24 February 2025 (secondary: Wikipedia 2025 in AI) and became generally available in May 2025. By 12 February 2026 its run-rate revenue was "over $2.5 billion," more than double the start-of-year level, with enterprises over half of that revenue and business subscriptions quadrupled since January (Series G announcement). No newer official figure was found. In the month to 10 October 2026 the npm package had 56.3M downloads, against 98.0M for OpenAI's Codex CLI and 1.74M for Gemini CLI. These include CI installs, so they are not user counts (Claude Code downloads, Codex downloads). The repository has 150.2k GitHub stars, more than Codex (128.7k) or Gemini CLI (107.3k), but Claude Code is proprietary, not open source. Its LICENSE.md reads "© Anthropic PBC. All rights reserved. Use is subject to Anthropic's Commercial Terms of Service," and the npm package's licence field points to the README (Claude Code repo, re-read 11 October 2026; npm registry: Claude Code). Embedding it in products offered to customers is governed by those terms.

The Agent SDK's Python repository has only 8,249 stars (about 30.0k for OpenAI's Agents SDK) and v0.2.165 from 8 October 2026 (GitHub API, PyPI). The TypeScript package drew 51.7M npm downloads in the last month, against 8.93M for @openai/agents (npm registry). Its licensing has two layers. The LICENSE file and the PyPI classifier say MIT, but the README says use "is governed by Anthropic's Commercial Terms of Service, including when you use it to power products and services that you make available to your own customers and end users," except for components with their own licence (Agent SDK repo, re-read 11 October 2026). Treat the SDK code as MIT and its use with Claude as bound by the Commercial Terms. Opus serves 54% of Claude Code sessions, against 10% of chat and Cowork sessions, so coding agents lean on the expensive tier (Economic Index report 6). Wikipedia names Boris Cherny as head of Claude Code, but it is a single weak source (Wikipedia: Boris Cherny).

Vendor tables put Opus 5.5 first, and independent indices mostly agree

Anthropic's launch pages report a consistent set of agentic benchmarks for its 5.x models. Every figure is vendor-reported except the tbench.ai board figures added to the Terminal-Bench row (third-party board, 330 trials per row, runs dated 1 Sep-6 Oct 2026). The GPT-6 Astra figures appear on Anthropic's page "as reported by OpenAI," which makes them second-hand.

Sources: Opus 5.5, Sonnet 5.5, Haiku 5.5, tbench.ai.

Independent OSWorld 2.0 results are far lower: the June 2026 paper's best was Claude Opus 4.8 (max thinking) at 20.6% binary completion (54.8% partial) at 500 steps, and later leaderboard rows (Epoch's copy of osworld-v2.xlang.ai, secondary) give Opus 5 (Max) 31.4% binary / 68.3% partial, with no Opus 5.5, Fable 5.1 or Sonnet 5.5 rows (arXiv 2606.29537).

These tables are not a leaderboard. Sonnet 5.5's vendor-reported 70.6% on Terminal-Bench 4.0 is not reproduced independently: the tbench.ai board gives Opus 5.5 64.85% and Sonnet 5.5 61.82%, both at max effort in Claude Code. Artificial Analysis's own Terminal-Bench 4.0 run orders them the other way (Sonnet 5.5 63.6%, Opus 5.5 59.6%), a separate measurement (tbench.ai; Artificial Analysis). The Opus page's prose gives FrontierCode 54.6% and CursorBench 52.5% at medium effort, while its table shows 54.4% and 57.8%. OSWorld appears as "partial," "offline subset" and, on the Fable page, version 2.0. AutomationBench counts safeguard interventions as failures, and Fable models scored zero on OSWorld 2.0 tasks where safeguards intervened (Opus 5.5, Fable). Anthropic itself says "benchmark margins have become a less reliable guide to real-world differences" (Opus 5.5). One comparison matters for cheap automation: Haiku 4.5 scored 0.0% on Terminal-Bench 4.0 and 15.7% on the OSWorld offline subset, against 39.2% and 72.4% for Haiku 5.5. For the first time, the cheapest tier is usable for light computer-use and agent steps (Haiku 5.5).

Independent measures broadly agree. Epoch AI placed Opus 5.5 top of its Epoch Capabilities Index on 1 October 2026 at 167, "narrowly ahead of GPT-6 Astra," with Sonnet 5.5 roughly level with Fable 5.1 at 165 (Epoch benchmarks). Epoch's 14 September snapshot showed Fable 5.1 at 164 (an unexplained one-point difference) and GPT-6 Astra at 166 (Epoch data insights). Epoch's own FrontierMath runs (updated 7 October 2026; small problem counts, indicative) give, on Tiers 1-3, Opus 5.5 91.2%, Fable 5.1 90.2%, Sonnet 5.5 88.8% and Haiku 5.5 75.1%, against 93.7% for GPT-6.1 Sol and GPT-6 Astra (Epoch benchmark data). Artificial Analysis's own Intelligence Index (v4.3.2, live 12 October 2026) gives Opus 5.5 (Max) 57.6, Sonnet 5.5 56.0, Fable 5.1 53.4 and GPT-6 Astra 52.7; Haiku 5.5 scores 43.4, near the open-weight GLM-5.3 (44.8). Its components put Opus 5.5 at 0.683 on normalised GDPval-AA (Sonnet 5.5 0.669, Fable 5.1 0.628, Haiku 5.5 0.559) and 61.4% on HLE without tools; median output speeds are 235 tokens a second (Haiku 5.5), 137 (Sonnet 5.5), 97 (Opus 5.5) and 65 (Fable 5.1) (Artificial Analysis). Arena's "Top 10 Agents" list, whose metric is undefined, ranks Opus 5.5 first (Arena). On LMArena's 8 October 2026 snapshot (preference Elo), Opus 5.5 is #1 on WebDev (1813 ±15; Sonnet 5.5 #3, Fable 5.1 #5, Haiku 5.5 #30) and #2 on text (1507 ±8), behind Gemini 4 Argon as a preliminary #1 (1525); on text Fable 5.1 is #6, Sonnet 5.5 #32, and Haiku 5.5 is not yet listed (LMArena text; WebDev).

The gaps are real. The 5.x pages drop SWE-bench and tau-bench entirely. The official SWE-bench file tops out at 79.2% for "Sonar Foundation Agent + Claude 4.5 Opus," and aggregator claims of 95% to 96% are not on it (SWE-bench leaderboards.json). Its newest Verified entries are from 26 February 2026, so the board is stale. Epoch's own Verified runs top out at 83.5% (Claude Opus 4.7, max, April 2026) with no 5.x rows, and SWE-bench Pro has no 2026 Claude 5.x rows (Epoch benchmark data; SWE-bench Pro). METR's pre-deployment review calls Opus 5.5 "an incremental improvement above Fable 5.1 … rather than a discontinuous jump" and "unlikely to be able to fully automate AI R&D." A preliminary METR assessment estimated "~1.5X overall acceleration in capabilities due to AI" inside Anthropic. METR's work was unpaid, and Anthropic reviewed the summary before sign-off (METR on Opus 5.5). METR's time-horizon page is no longer updated (METR time horizons). No METR time-horizon value exists for Opus 5.5, Fable 5.1 or Sonnet 5.5. Compare vendors in Top 10 models compared, OpenAI and Gemini.

Run-rate revenue rose from $14 billion to a reported $65 billion in five months

Anthropic's own disclosures run to May 2026. The July figure comes from press reports only.

Anthropic's Series H page (re-read 11 October 2026) names Capital Group, Coatue, D1, GIC, ICONIQ and XN as co-leads and includes $15B of previously committed hyperscaler investment, $5B of it from Amazon (Anthropic Series H). Wikipedia's Reuters-based account omits Greenoaks (Reuters, headline only). Anthropic's 6 April compute post gives about $9B at end-2025 and above $30B in April (Anthropic), and secondary aggregators match these and the $47B May figure (Unite.ai). Run-rate grew about 3.4x between February and May 2026 on Anthropic's own figures, and about 4.6x to July if the press figure holds. Use ranges, not point values.

Other scale markers are thinner. Customers spending over $100,000 a year grew 7x in the year to February 2026 (Series G page), and Wikipedia lists about 2,500 employees (Wikipedia). Secondary summaries of Menlo Ventures data put Anthropic at 40% of enterprise LLM API spend in late 2025, against 27% for OpenAI and 21% for Google (usecarly summary, Longterm Wiki). The mid-2025 baseline is either 32% (BigDATAwire) or 42%, and the report itself was unreachable. On OpenRouter, Anthropic had only 2.5% of requests in the week from 28 September 2026, behind DeepSeek (21.8%), Google (21.5%) and OpenAI (18.2%), yet Claude models headed the one OpenRouter per-task share-of-spend list captured (its "top models by task" section; the task was not recorded) (OpenRouter rankings). Neither OpenRouter figure measures users or overall spend.

Named deployments cluster in finance and consulting. Barclays (1 October 2026) is targeting Claude Code use by half its developers by end-2026. It runs a knowledge assistant for 16,000+ colleagues serving 20M+ UK retail customers and routes about 120,000 client emails a day, with no savings published (Barclays). The Claude Frontier Academy (2 October) commits $100M to train 10,000 "Frontier Deployed Engineers" by end-2027. Its cohort includes Accenture, Bain, Capgemini, Deloitte, McKinsey and Morgan Stanley, and Anthropic counts 175,000+ Claude certifications across 46,000 firms (Frontier Academy). The newsroom lists an Accenture partnership on 18 September (Anthropic news), and Salesforce announced "Claudeforce" on 26 August (title only: Salesforce). Two acquisitions are known from titles only: Bun, in December 2025, under a URL citing a Claude Code "$1B milestone" (Anthropic), and the computer-use company Vercept, in February 2026 (Anthropic).

The leadership page lists Dario Amodei (CEO), Daniela Amodei (President and board chair), Jared Kaplan (Chief Science Officer), Jack Clark (Head of Public Benefit), Krishna Rao (CFO), Rahul Patil (CTO), Paul Smith (Chief Commercial Officer) and Ami Vora (Chief Product Officer) (leadership). Former Chief Product Officer Mike Krieger moved to Anthropic Labs on 13 January 2026 (Introducing Labs). Andrej Karpathy's move to Anthropic was reported in an Axios headline of 19 May 2026 (single-source: seen through the Hacker News index; the article was not read, and no team was confirmed) (Axios headline). Amodei's September 2026 essay argues "we must slow the pace at which we improve the capabilities of AI models," while "pacing does not mean halting model training" (We Must Pace the Frontier). See People and Companies.

The Economic Index shows the API automates while chat augments

Anthropic's Economic Index classifies samples of Claude conversations (about 1 million in November 2025) by occupation and task. It separates "automation" (Claude does the task) from "augmentation" (Claude helps a person do it). The Hugging Face dataset has six releases, from 10 February 2025 to 26 June 2026. The data is CC-BY and the code MIT (dataset card). The figures are Anthropic's own classifier output, not ground truth.

Sources: report 4, India brief, report 5, report 6.

The June report's survey, which Anthropic says is "not representative of the general population," found 86% of respondents reporting speed gains, 82% scope gains and 69% quality gains. Women were 12% of the linked sample (report 6). A secondary article's claims that 93% of conversations produce a tangible output and that higher-wage tasks use 2.07x more tokens were not found in the primary text (blockchain.news, snippet). The 2.07x is close to 1.34 × 1.53 ≈ 2.05, but quote the primary ratios. An Economic Index connector launched on 22 July 2026 (snippet only: Anthropic).

Two implications follow. API-led agents sit in the segment Anthropic calls automation-dominant. Higher-wage knowledge work takes more turns and output, so analyst, legal and finance automations should be budgeted per task from logged token counts, not estimated from flat chatbot fees. See use cases.

India is Claude's second-largest market but ranks 101st per capita

Anthropic opened its Bengaluru office on 16 February 2026, in the week of the India AI Impact Summit in New Delhi. It is led by Managing Director Irina Ghose and is Anthropic's second Asian office after Tokyo. Named partners include Air India (Claude Code for developers), CRED (claiming "2x faster feature delivery and 10% better test coverage," a vendor claim), Cognizant ("deploying Claude to 350,000 employees"), Razorpay, Pratham, Adalat AI and Digital Green. Anthropic also cites a government MCP server for the Ministry of Statistics (MoSPI) and Indic-language work in ten languages, from Hindi and Bengali to Malayalam and Urdu (Bengaluru office). The same post calls India "the second-largest market for Claude.ai," a company statement that Fortune also reported from the summit (Fortune).

India had 5.8% of global Claude.ai consumer use in November 2025, second only to the US, but ranked 101st of 116 countries per capita. Maharashtra (15.5%), Tamil Nadu (13.2%), Karnataka (12.7%) and Delhi (10.5%) account for over half of Indian use. Software tasks are 45.2% of mapped Indian tasks, the highest share of any country. 51.3% of Indian use is work-related (46% globally). Indian users finish in 14.8 minutes tasks that would take 3.8 hours unaided, a 15x speed-up against 12x globally, and delegate more (3.60 against 3.38 on a 1-to-5 autonomy scale) (India brief). The brief covers consumer Free, Pro and Max users located by IP, not business API use. The September 2025 report gave India 7.2% of global use and a per-capita index of 0.27 against 3.62 for the US, from an earlier period and only partly re-checked (Economic Index, Sept 2025). The June 2026 release (April and May 2026 data; Claude.ai chat and Cowork on Free, Pro and Max plans only) gives India 6.92% and 7.12% of all usage and a usage-per-capita index of 0.31 and 0.30 (1.0 = proportional to working-age population), against 4.25 and 3.87 for the US (Economic Index dataset, file release_2026_06_26/data/aei_claude_ai_2026-06-26.csv). Releases cover different periods, so ranks are not recomputed here.

India is on Anthropic's supported countries list, but the API has no INR price list or stated GST treatment (rupee consumer and Team prices, GST included, are reported from 13 July 2026 by one outlet), the first-party API has no India inference region (India-only routing exists only for Opus 5, Sonnet 5 and Haiku 4.5 on Bedrock), and there is no Indian enterprise customer count. The Frontier Academy announcement does not mention India. Use is concentrated in IT-heavy states and software tasks, so the per-capita gap points to SMEs outside that corridor as the least-served users. See India and Asia-Pacific.

Compliance covers SOC 2 and ISO 27001, while the newest models carry retention strings

Anthropic's regional compliance page lists SOC 2 Type 2, ISO/IEC 27001, 27017 and 27018, HIPAA and GDPR with EU/EEA processing. It does not list ISO/IEC 42001 or FedRAMP (regional compliance). The Trust Center returned no readable content, so SOC 2 report access and BAA scope could not be checked. Enterprise adds SCIM, audit logs, a Compliance API, custom retention and a "HIPAA-ready offering" (claude.com pricing). In Claude's organisational memory, owners enable it, members opt in, and owners cannot read individual memories. Memory is unavailable to HIPAA, public-sector and custom-retention organisations (Claude memory help), and regulated buyers may reject memory their administrators cannot inspect.

Anthropic's safety framework now shapes what customers can buy. The Responsible Scaling Policy is at version 3.4 (effective 8 July 2026), with Risk Reports published since version 3.0 and no mapping of current models to ASL levels (RSP). Mythos 5.1 is limited to verified organisations, and on 6 October 2026 the Cyber Verification Program was split into three access tiers covering Opus 5.5, Sonnet 5.5 and Mythos 5.1 (Cyber Verification Program). A Life Sciences Verification Program followed on 17 September (Anthropic news), and Project Glasswing extended Mythos access to about 150 more organisations in 15+ countries on 2 June (Mythos). Fable-class models require 30-day retention. Enterprise Frontier Safeguards will pair customer-controlled storage with misuse detection (Enterprise Frontier Safeguards). A documented "refusals and fallback" pattern retries on another Claude model when a classifier blocks a request (Fable 5.1 model page).

The 2026 Usage Policy update (8 October, effective 12 November 2026) adds rules on deceptive campaigns, autonomous physical actions and high-risk health and finance uses (usage policy update). Behind the controls: in November 2025 Anthropic reported that a group it assessed as Chinese state-sponsored used Claude Code against about 30 organisations, with AI doing 80% to 90% of the work (AI espionage report). In February 2026 it accused DeepSeek, Moonshot AI and MiniMax of distillation attacks using 16M+ chats (secondary: Wikipedia). The authors' copyright settlement (Bartz v. Anthropic, at least $1.5B per Wikipedia) received final court approval on 20 July 2026, with first payments expected by 15 November 2026 (settlement website). See governance and security.

2026 proved that frontier-model access is a regulatory risk

Two events matter more to a buyer than any benchmark. At 5:21pm ET on 12 June 2026, a US export-control directive suspended all access to Fable 5 and Mythos 5 by any foreign national, including Anthropic's own staff, and Anthropic disabled both models for every customer. Anthropic said the concern was a jailbreak demonstration that found "previously known, minor vulnerabilities," and other models were unaffected (Fable and Mythos access statement). Mythos went to "trusted" US organisations on 26 June (headline only: Semafor). Commerce removed the controls on 30 June (Al Jazeera), and access returned on 1 July (Fable 5 and Mythos 5). Indian users of Fable 5 lost it for about 19 days.

The second is the Pentagon dispute, documented unevenly. The supply-chain-risk designation was directed on 27 February and formalised on 5 March, and a federal judge blocked it on 26 March, all from headline listings (WSJ, CNN). An appeals court upheld it on 25 September (headline only: CNBC). On 5 October the BBC reported that the US Department of Defense "has ceased the use of Anthropic products," including removal from Palantir's Maven Smart System. Hegseth had set a six-month phase-out on 27 February, to end by late August, yet BBC sources said Claude was still in use "as recently as last week," and the delay is unexplained (BBC, re-read 11 October 2026). Reuters, via the BBC, reported that Anthropic plans to warn potential IPO investors that AI may pose "catastrophic or existential risks to humanity" (BBC).

Access to the most capable tier now depends on governments, verification programmes and the vendor's classifiers, not only on price. See the timeline and future outlook.

What this means for your business

Claude is a family of tiers, not one model. If you are weighing it for automation, choose the tier for each workflow step, use the discounts deliberately, and settle residency, retention and fallback before any workflow depends on it.

Match the model to the job

What to budget

Tokens cost from fractions of a cent to about $5 per task (above); the cost research multiplies by 1.2 to 2 for a production budget. Cache first: agent loops resend the same prompt, tools and documents every turn, and cached reads on Opus 5.5 and Sonnet 5.5 cost 5% of base input. Batch anything that can wait (50% off). Route routine steps to Haiku 5.5, since Sonnet 5.5 input costs 20x and Opus 5.5 40x as much, and keep Haiku prompts under 100K tokens. Allow for the ~30% tokenizer increase, 10% for US-only processing and $0.08 (₹7.68) per Managed Agents session-hour. Per-outcome SaaS fees of $0.99 to $2.00 (₹95 to ₹192) are the build-or-buy reference. API prices are USD-only, so INR budgets carry exchange-rate risk: the X-Rates monthly average rose from 90.73 in January to 95.95 in October 2026 (X-Rates). See AI automation costs.

Data handling and India processing

• US: inference_geo: "us" (1.1x), Bedrock's US endpoints or profile, Google Cloud's US multi-region (+10%) or Foundry's US Data Zone Standard. If you sell to the US Department of Defense, first confirm whether Claude is acceptable at all.
• India: the first-party API processes globally or in the US, and no cloud documents India processing for the 5.5, Fable or Mythos models. Bedrock's in. profiles keep Opus 5, Sonnet 5 and Haiku 4.5 in Mumbai and Hyderabad, so India-only processing today means the previous generation (AWS regional availability); confirm with GetInferenceProfile.
• Retention: ZDR is available for Opus 5.5 with Anthropic's approval but excludes Batch, the Files API, code execution, the MCP connector, Agent Skills, Managed Agents and Fable/Mythos-class models (30-day retention). Managed Agents memory stays with Anthropic, so keep regulated data on the Messages API.
• Health data: the Enterprise HIPAA-ready offering, with a BAA whose scope you confirm; Managed Agents has no BAA.

Deadlines: Usage Policy update 12 November 2026; DPDP consent managers 13 November 2026; Sonnet 4.5 retirement 30 November 2026; DPDP core duties (72-hour breach reports, one-year logs, cross-border conditions) 13 May 2027, by which data-processing terms should state processing location and retention.

Risks to plan for

The likeliest failures are a spend cap, a retirement, a breaking change, a classifier refusal or a government action, not a wrong answer. Use separate workspaces per team or workflow, with spend alerts below the cap. Configure Anthropic's refusal fallback to another Claude model, and keep a tested fallback outside Anthropic: the June 2026 export-control action took a model family offline for 19 days. DeepSeek documents an Anthropic-compatible endpoint (DeepSeek news), though China hosting raises its own residency questions; compare DeepSeek, OpenAI, Gemini and Grok. Run computer-use and coding agents in containers with restricted egress and no credentials, and require human approval for payments, deletions and outbound messages.

What to test before relying on it

• Your own tasks, not vendor tables. Sonnet 5.5's vendor-reported 70.6% on Terminal-Bench 4.0 is 61.82% on the tbench.ai board, and independent OSWorld 2.0 scores (best 31.4% binary) sit far below vendor figures. Define evaluations before scaling (Building effective agents, Demystifying evals, context engineering).
• Cost on the exact model ID, with real prompts near Haiku's 100K step.
• Breaking changes: forced tool choice, Haiku 5.5 sampling parameters and older computer-use tools can return 400 errors.
• Spend-cap behaviour: at the cap, every request gets HTTP 429 until the 1st of the next month.
• Both fallbacks end to end, and prompt injection on agents that read untrusted content: a researcher bypassed Claude Code's default auto mode about 80% of the time.

Questions to ask any vendor or implementation partner

• Which model IDs and route (Claude API, Bedrock, Google Cloud or Foundry) does each workflow use, and where is it processed?
• Which features fall outside ZDR or a BAA, and what is retained, for how long and by whom?
• Which spend tier applies, and how is one workflow kept from halting the others?
• What is the tested fallback outside Anthropic, and how are upgrades regression-tested?
• What results on tasks like yours support the tier chosen? Anthropic counts 175,000+ Claude certifications and runs Anthropic Academy courses, but no formal exam was found (Anthropic Academy), so ask for results, not certificates. The playbook covers implementation in more detail.

Open questions and unverified items

Sources

1. Claude Platform: Models overview
2. Claude Platform: Release notes
3. Claude Platform: Pricing
4. Claude Platform: Rate limits
5. Claude Platform: Data residency
6. Claude Platform: Fable 5.1 model page
7. Claude plans and pricing (claude.com)
8. Anthropic: Claude Opus 5.5
9. Anthropic: Claude Sonnet 5.5
10. Anthropic: Claude Haiku 5.5
11. Anthropic: Claude Fable
12. Anthropic: Claude Mythos
13. Anthropic: Claude Opus 5 announcement
14. Anthropic: Claude Fable 5 and Mythos 5
15. Anthropic: Fable and Mythos access statement (12 Jun 2026)
16. Al Jazeera: US lifts restrictions on Fable and Mythos (1 Jul 2026)
17. Semafor: US releases Mythos to some US companies (headline only)
18. Claude Platform: Managed Agents overview
19. Claude Platform: Computer use tool
20. Claude Platform: Programmatic tool calling
21. Claude Platform: Tool search tool
22. Claude Platform: Memory tool
23. Claude Platform: Agent Skills overview
24. Claude Code docs: Agent SDK overview
25. Claude Code docs: GitHub Actions
26. Claude in Chrome
27. Anthropic: Donating MCP and establishing the Agentic AI Foundation
28. MCP blog
29. MCP blog: Enterprise-Managed Authorization
30. Anthropic: Acquiring Stainless
31. Anthropic engineering: Building effective agents
32. Anthropic engineering: Demystifying evals for AI agents
33. Anthropic engineering: Effective context engineering for AI agents
34. Simon Willison: prompt-injection tag
35. Invariant Labs: GitHub MCP vulnerability
36. Invariant Labs: MCP tool poisoning attacks
37. Wikipedia: 2025 in artificial intelligence (secondary)
38. npm downloads: @anthropic-ai/claude-code
39. npm downloads: @openai/codex
40. GitHub: anthropics/claude-code
41. GitHub API: anthropics/claude-agent-sdk-python
42. PyPI: claude-agent-sdk
43. npm registry: @anthropic-ai/claude-agent-sdk
44. Wikipedia: Boris Cherny (secondary, weak)
45. Epoch AI: Benchmarks hub
46. Epoch AI: Data insights
47. OpenRouter: Models API
48. OpenRouter: Rankings
49. Arena leaderboard
50. SWE-bench official leaderboards.json
51. METR: Claude Opus 5.5 pre-deployment evaluation
52. METR: Time horizons
53. OpenAI API pricing
54. Intercom pricing
55. Zendesk pricing
56. X-Rates: USD to INR monthly average, 2026
57. Claude status page: incident history
58. Claude: Regional compliance
59. Anthropic: Enterprise Frontier Safeguards
60. Claude Help Center: Chat search and memory
61. Anthropic Commercial Terms
62. Gazette of India: G.S.R. 843(E), DPDP commencement
63. Gazette of India: G.S.R. 846(E), DPDP Rules 2025
64. Anthropic: Series E announcement
65. Anthropic: Series G announcement
66. Anthropic: Series H
67. Reuters: Anthropic raises $65 billion, valued at $965 billion (headline only)
68. Anthropic: Confidential draft S-1
69. Wikipedia: Anthropic (secondary)
70. Hacker News: Bloomberg, "Anthropic Nears $20 Billion Revenue Run Rate"
71. Hacker News: CNBC, run rate climbed to $65B in July
72. Hacker News: Bloomberg, run rate surpasses $65B
73. Hacker News: Reuters, run rate tops $65 billion
74. Yahoo Finance: Anthropic aims to nearly triple revenue (snippet)
75. Unite.ai: Anthropic run-rate coverage (secondary)
76. usecarly: LLM market share 2026 (secondary)
77. Longterm Wiki: sourcing page (secondary)
78. BigDATAwire: Menlo Ventures enterprise LLM spend (secondary)
79. Anthropic: Barclays scales Claude
80. Anthropic: Claude Frontier Academy
81. Anthropic newsroom
82. Salesforce: Salesforce and Anthropic announce Claudeforce (title only)
83. Anthropic: Acquires Bun (title only)
84. Anthropic: Acquires Vercept (title only)
85. Anthropic: Leadership
86. Anthropic: Introducing Labs
87. Axios headline via Hacker News index (19 May 2026; article not read)
88. Dario Amodei: We Must Pace the Frontier
89. Hugging Face: Anthropic Economic Index dataset
90. Anthropic Economic Index: January 2026 report
91. Anthropic Economic Index: March 2026 report
92. Anthropic Economic Index: June 2026 report
93. Anthropic Economic Index: September 2025 report
94. blockchain.news: June 2026 Economic Index coverage (snippet)
95. Anthropic: Economic Index connector (snippet)
96. Anthropic: India Country Brief
97. Anthropic: Bengaluru office and partnerships across India
98. Fortune: India AI Impact Summit closes (23 Feb 2026)
99. Anthropic: Supported countries and regions
100. Anthropic Trust Center
101. Anthropic: Responsible Scaling Policy
102. Anthropic: Expanding the Cyber Verification Program
103. Anthropic: 2026 Usage Policy update
104. Anthropic: Disrupting AI-orchestrated espionage
105. WSJ: Pentagon formally labels Anthropic supply-chain risk (headline only)
106. CNN: Judge blocks Pentagon effort against Anthropic (headline only)
107. CNBC: Appeals court upholds Pentagon designation (headline only)
108. BBC: US Department of Defense ceases use of Anthropic products
109. BBC: OpenAI withholds GPT-6.1 Astra; Anthropic IPO risk warning
110. Axios: Claude Opus 4.6 software hunting (headline only)
111. Anthropic: Cyber Mission
112. Anthropic: Genesis Mission commitment
113. DeepSeek API news
114. Anthropic Academy (Skilljar)
115. Claude Platform: Claude in Amazon Bedrock (regions)
116. Claude Platform: Claude on Google Cloud (endpoints)
117. Anthropic Privacy Center: Zero data retention scope
118. npm registry: @anthropic-ai/claude-code
119. Anthropic Copyright Settlement website
120. GitHub: anthropics/claude-agent-sdk-python (README and LICENSE)
121. AWS: Amazon Bedrock model support by Region
122. AWS: Bedrock model card, Claude Sonnet 5
123. Google Cloud: Gemini Enterprise Agent Platform data residency
124. Microsoft Learn: Deployment types for Foundry Models
125. Claude Platform: API and data retention
126. Claude Platform: Extended thinking
127. Claude Platform: Service tiers
128. TechStory: Anthropic launches rupee billing for Claude in India (single-source)
129. Linux Foundation: Formation of the Agentic AI Foundation
130. Official MCP Registry API
131. tbench.ai: Terminal-Bench 4.0 board
132. Artificial Analysis: Models leaderboard
133. LMArena: Text leaderboard
134. LMArena: WebDev leaderboard
135. arXiv 2606.29537: OSWorld 2.0
136. Epoch AI: Benchmark data (bulk download)
137. Scale: SWE-bench Pro leaderboard
138. Anthropic: Google and Broadcom compute partnership (6 Apr 2026)