Anthropic released Claude Opus 5.5 on September 22, 2026, as the first model in its Claude 5.5 family. The company says it performs roughly at Claude Fable 5.1’s level on most work while costing 40% less to run than Opus 5. Opus 5.5 arrived two months after Opus 5, released July 24, 2026; Anthropic also confirmed that Sonnet 5.5 and Haiku 5.5 would launch soon. Anthropic called Opus 5.5 its strongest model yet in internal alignment testing, while cautioning that benchmark margins are becoming a less reliable guide to real-world quality.

The company listed stronger agentic coding and knowledge work, output generation more than 30% faster, clearer writing and a 40% lower typical cost as changes from Opus 5. In independent Artificial Analysis testing, Opus 5.5 scored 58 on the Intelligence Index, an aggregate of ten evaluations. Its output speed ranged from 74 to 86 tokens per second, while cost per index task ranged from $0.55 at low effort to $5.98 at maximum effort, an 11x spread.

Anthropic cited an early tester’s migration of 680,000 lines of code in under a day and another tester’s audit and repair of a 200,000-line codebase in under three hours. For the latter task, Opus 5 took over 20 hours and used 2.5 times the tokens. In an internal C-to-Rust translation of HAProxy, Opus 5.5 and Fable 5.1 passed nearly all regression tests; Opus 5.5 finished in 9.5 hours versus 12 hours, at 51% lower cost. Anthropic also said Opus 5.5 beat GPT-6 Astra on FrontierCode at roughly one-fifth the cost per task, matched it on Terminal-Bench 4.0 at about 40% of the cost, and beat GPT-5.6 Sol on CursorBench by 11 points at about one-third the cost.

Company-reported coding results included Clio’s unattended run of over 18 hours across six repositories, Quantium reducing a task from 38 prompts over four days to 11 prompts over three hours, and Optiver matching Opus 5’s quality in about half the turns, time and tokens. Kiro reported more tasks solved than Opus 5 on a public benchmark using about 40% fewer calls and half the tokens. For knowledge work, Opus 5.5 met a quality bar in 16 of 18 attempts at writing an earnings report from difficult-to-find source material; Fable 5.1 and Opus 5 did not pass once. Hebbia said it covered 86.6% of an expert rubric, compared with 60.3% for Opus 5, and Deloitte Consulting reported catching 72% of known code review bugs at Opus 5.5’s lowest setting, versus 56% for Opus 5 at its highest.

Anthropic said the model leads with key information, uses less jargon and follows custom writing instructions more consistently. Stripe reported that Opus 5.5 handled a 40-pull-request rebase across a dozen sessions, with all 40 passing continuous integration the next afternoon. Box reported answers 40% less verbose while using one-third the tokens, with no loss in accuracy; Factory said it matched Opus 5’s high-effort quality using 20 to 25% fewer output tokens. Batch API use receives a 50% discount on input and output tokens. Fast mode, available on Claude Code and Claude Platform, runs up to 2.5x faster and costs $8 per million input tokens and $40 per million output tokens.

The model has a 1M-token context window and a maximum output of 128K tokens, or 300K on the beta Batch API. Its knowledge cutoff is June 2026; adaptive thinking is always on, default effort is medium, and comparative latency is listed as moderate. The model identifier is claude-opus-5-5 across the Claude API, Google Cloud, Microsoft Foundry and Claude Platform on AWS; Amazon Bedrock uses anthropic.claude-opus-5-5. Four breaking changes affect Opus 5 integrations: thinking cannot be disabled, forced tool use returns an error, thinking blocks are tied to the model and conversation that produced them, and the older computer_20251124 tool is no longer accepted on the Claude API or Google Cloud. Anthropic also raised five-hour usage limits on Pro, Max, Team and seat-based Enterprise plans, and added a saveable rate-limit reset for subscription users.

Opus 5.5 is available on the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS. Anthropic said it would keep the model active for at least a year, with retirement no earlier than September 22, 2027. External evaluators METR, Frontier Design and the US Center for AI Standards and Innovation tested it before release. Anthropic reported that its automated behavioral audit showed improvement over prior Claude models on nearly every measure;,

Anthropic released Claude Opus 5.5 on September 22, 2026, as the first model in its Claude 5.5 family. The company says it performs roughly at Claude Fable 5.1’s level on most work while costing 40% less to run than Opus 5. Opus 5.5 arrived two months after Opus 5, released July 24, 2026; Anthropic also confirmed that Sonnet 5.5 and Haiku 5.5 would launch soon. Anthropic called Opus 5.5 its strongest model yet in internal alignment testing, while cautioning that benchmark margins are becoming a less reliable guide to real-world quality.

The company listed stronger agentic coding and knowledge work, output generation more than 30% faster, clearer writing and a 40% lower typical cost as changes from Opus 5. In independent Artificial Analysis testing, Opus 5.5 scored 58 on the Intelligence Index, an aggregate of ten evaluations. Its output speed ranged from 74 to 86 tokens per second, while cost per index task ranged from $0.55 at low effort to $5.98 at maximum effort, an 11x spread.

Anthropic cited an early tester’s migration of 680,000 lines of code in under a day and another tester’s audit and repair of a 200,000-line codebase in under three hours. For the latter task, Opus 5 took over 20 hours and used 2.5 times the tokens. In an internal C-to-Rust translation of HAProxy, Opus 5.5 and Fable 5.1 passed nearly all regression tests; Opus 5.5 finished in 9.5 hours versus 12 hours, at 51% lower cost. Anthropic also said Opus 5.5 beat GPT-6 Astra on FrontierCode at roughly one-fifth the cost per task, matched it on Terminal-Bench 4.0 at about 40% of the cost, and beat GPT-5.6 Sol on CursorBench by 11 points at about one-third the cost.

Company-reported coding results included Clio’s unattended run of over 18 hours across six repositories, Quantium reducing a task from 38 prompts over four days to 11 prompts over three hours, and Optiver matching Opus 5’s quality in about half the turns, time and tokens. Kiro reported more tasks solved than Opus 5 on a public benchmark using about 40% fewer calls and half the tokens. For knowledge work, Opus 5.5 met a quality bar in 16 of 18 attempts at writing an earnings report from difficult-to-find source material; Fable 5.1 and Opus 5 did not pass once. Hebbia said it covered 86.6% of an expert rubric, compared with 60.3% for Opus 5, and Deloitte Consulting reported catching 72% of known code review bugs at Opus 5.5’s lowest setting, versus 56% for Opus 5 at its highest.

Anthropic said the model leads with key information, uses less jargon and follows custom writing instructions more consistently. Stripe reported that Opus 5.5 handled a 40-pull-request rebase across a dozen sessions, with all 40 passing continuous integration the next afternoon. Box reported answers 40% less verbose while using one-third the tokens, with no loss in accuracy; Factory said it matched Opus 5’s high-effort quality using 20 to 25% fewer output tokens. Batch API use receives a 50% discount on input and output tokens. Fast mode, available on Claude Code and Claude Platform, runs up to 2.5x faster and costs $8 per million input tokens and $40 per million output tokens.

The model has a 1M-token context window and a maximum output of 128K tokens, or 300K on the beta Batch API. Its knowledge cutoff is June 2026; adaptive thinking is always on, default effort is medium, and comparative latency is listed as moderate. The model identifier is claude-opus-5-5 across the Claude API, Google Cloud, Microsoft Foundry and Claude Platform on AWS; Amazon Bedrock uses anthropic.claude-opus-5-5. Four breaking changes affect Opus 5 integrations: thinking cannot be disabled, forced tool use returns an error, thinking blocks are tied to the model and conversation that produced them, and the older computer_20251124 tool is no longer accepted on the Claude API or Google Cloud. Anthropic also raised five-hour usage limits on Pro, Max, Team and seat-based Enterprise plans, and added a saveable rate-limit reset for subscription users.

Opus 5.5 is available on the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS. Anthropic said it would keep the model active for at least a year, with retirement no earlier than September 22, 2027. External evaluators METR, Frontier Design and the US Center for AI Standards and Innovation tested it before release. Anthropic reported that its automated behavioral audit showed improvement over prior Claude models on nearly every measure;,