Claude Opus 5.5 Launches With 1M Context and Lower Pricing

Anthropic released Claude Opus 5.5 on September 22, 2026, opening a new Claude 5.5 model family rather than shipping a minor point update. Anthropic positions the model for long-running agentic coding and knowledge work, and calls it a major step up from Opus 5 on performance and efficiency. Documentation and API access are live on the Claude Platform and across major cloud providers.

Release date and availability span consumer and cloud tiers

Claude Opus 5.5 launched on September 22, 2026. It is available to Claude users on the Pro, Max, Team, and Enterprise plans, and through the Claude Platform, Amazon Bedrock, Google Cloud, and Microsoft Foundry. Developers call it through the Claude API using the model identifier claude-opus-5-5.

Access depends on having an account, credentials, and applicable billing or credits, as with any API model. Availability is not restricted to a single "enterprise and developer" tier, and it is not automatically enabled by subscription alone.

Opus 5.5 is a major step up, not an incremental patch

Anthropic frames Opus 5.5 as a significant advance over Opus 5, centered on agentic work rather than a narrow reasoning tweak. The company highlights faster output, lower token consumption, clearer communication, better adherence to user-specified writing rules, and stronger safety measures.

The headline efficiency claims: Opus 5.5 runs roughly 40% cheaper than Opus 5 on typical workloads, combining lower per-token prices with improved token efficiency. Anthropic also reports more than 30% faster output generation than Opus 5. On most work, Anthropic says Opus 5.5 performs at the level of Fable 5.1.

Safety received direct attention. Anthropic cites greater resistance to prompt injection along with additional cybersecurity and biology safeguards.

Anthropic published a full benchmark table at launch

Anthropic disclosed benchmark results across coding, knowledge work, computer use, and multidisciplinary reasoning. Reported suites include Terminal-Bench 4.0, FrontierCode v1.1, CursorBench 4.0, GDPval-AA v2.1, AutomationBench, Humanity's Last Exam, Terminal-Bench-Science, OSWorld 2.0, and Chartography.

Selected vendor-reported figures, comparing Opus 5.5 against Opus 5 under Anthropic's stated conditions:

Vendor-reported results from Anthropic. Preserve effort level and tool availability when citing.

These are Anthropic's own numbers rather than independently verified scores, and the conditions (effort level, tool access) matter to interpretation. Anthropic's published comparison table also names GPT-6 Astra and GPT-5.6 Sol alongside Fable 5.1 and Opus 5, so any framing against other vendors should reflect that set rather than substitute a different lineup.

These are Anthropic's own numbers rather than independently verified scores our Claude Opus 5.5 benchmarks guide puts the two side by side, and the conditions.

Pricing dropped from the Opus 5 tier

Opus 5.5 costs less than Opus 5. Standard pricing is $4 per million input tokens and $20 per million output tokens, down from $5 and $25 for Opus 5.

Standard API pricing as of September 2026.

Anthropic also offers a research-preview fast mode for Opus 5.5 at $8 per million input tokens and $40 per million output tokens, plus Batch API pricing of $2 per million input tokens and $10 per million output tokens.

On positioning, Opus 5.5 sits above Sonnet 5 on price but below Fable 5.1. It is more expensive than Sonnet 5, and it is substantially cheaper than Fable 5.1.

The model fits between Sonnet 5 and Fable 5.1 in the lineup

Opus 5.5 joins Claude Opus 5, Claude Fable 5.1, and Claude Sonnet 5 in Anthropic's current lineup. Rather than a blanket "most capable model" claim, Anthropic describes Opus 5.5 as performing at the level of Fable 5.1 on most work while emphasizing specific workload strengths in agentic coding, computer use, and knowledge work.

Technical specifications:

• Context window: 1 million tokens by default.
• Maximum output: 128,000 tokens.
• Thinking: always-on adaptive thinking, with effort as the control rather than a manual on/off switch.

The 1 million-token context window is a central change. Earlier reporting that cited a 200,000-token window is incorrect for this model.

Existing integrations need migration, not a drop-in swap

Opus 5.5 is not a drop-in replacement. Anthropic's release notes document behavioral and compatibility changes that can break existing calls.

Key changes to review before switching:

• Thinking cannot be disabled. Sending thinking: {"type": "disabled"} or a manual thinking: {"type": "enabled", ...} returns HTTP 400.
• The tool_choice values any and tool return errors. Use auto, with strict tool use where appropriate.
• Computer-use workflows on the Claude API and Google Cloud must use computer_toolset_20260801. The prior computer_20251124 tool returns HTTP 400 on those platforms.
• Regression-test existing workloads for output style, tool calling, latency, token usage, and safety-routing behavior.

Anthropic links to migration guidance in its documentation, and teams should consult it before moving production traffic.

What the launch signals

Claude Opus 5.5 reflects Anthropic's push toward faster, cheaper agentic models rather than a slow cadence of monolithic releases. The efficiency story, lower price, faster output, and larger context, lowers the cost of adoption for teams already building on Claude. As independent benchmarks and real-world reports accumulate, the model's standing against GPT-6 Astra, GPT-5.6 Sol, and other frontier releases will come into clearer focus. For now, the launch pairs a 1 million-token context window with a price cut, aimed squarely at long-running coding and knowledge-work agents.