Skip to content
NewA calmer, faster ZeroHour
StoryModel releaseinfo

Anthropic launches Claude Opus 5.5 and Sonnet 5.5

4 sources · 4 articlesFirst reported Updated ()2
What's new

First coverage.

Merged summary · grok-4.7 · rewritten as coverage arrivesSummaries by Vani AI

Anthropic released cheaper Claude Opus 5.5 on 22 September with Fable-class safeguards, then Sonnet 5.5 and prompting guidance on 28 September.

Anthropic released Claude Opus 5.5 on 22 September 2026, the first Claude 5.5 model and its first launch after CEO Dario Amodei said the company would pace frontier development. Anthropic says it matches Claude Fable 5.1 on most work at about 40% lower operating cost than Opus 5, with list prices cut from $5/$25 to $4/$20 per million input/output tokens, cache reads at $0.20, and output more than 30% faster; Latent Space notes the list cut is 20% and that higher token use can erase savings at maximum effort. On Terminal-Bench 4.0 it scored 66.4%, above Fable 5.1 at 55.8% and GPT-6 Astra at 57.9%, with other reported figures of 54.4% on FrontierCode v1.1, 65.3% on FrontierCode 1.1 Extended, 57.8% on CursorBench, and 81.8% on OSWorld 2.0, while MarkTechPost says GPT-6 Astra still leads Terminal-Bench-Science and AutomationBench. Sources disagree on ranking: several say it matches Fable 5.1 on most tasks, TechCrunch says it beat larger Fable on many coding and knowledge benchmarks and is the strongest model Anthropic has tested, and The Verge limits "strongest" to Anthropic's broadest alignment test. Cyber and biology capability is described as comparable to Mythos 5.1, so Fable-class safeguards reroute risky requests—The Verge says cybersecurity goes to Opus 4.8 and flagged biology to Opus 5, Hacker News says Opus 4.8 or Opus 5, and The Decoder adds frontier-development routing—while TechCrunch says limits cover exploit discovery in compiled programs and biological-weapons assistance. Opus 5.5 became the default in Claude Code and the Claude app at medium effort, with always-on thinking and pre-release tests by METR and Frontier Design; OpenAI shipped GPT-6 Sol and Luna about an hour later, and on 28 September Anthropic published prompting guidance and released Sonnet 5.5 at $2/$10, reporting 70.6% on Terminal-Bench 4.0 and new Sonnet cyber safeguards.

Key facts
  • Claude Opus 5.5 launched 22 September 2026 as the first Claude 5.5 model; closed-weight API id claude-opus-5-5 on the Claude Platform, AWS, Google Cloud, and Azure.
  • List price fell from $5/$25 to $4/$20 per million input/output tokens, with cache reads at $0.20; Anthropic cites about 40% lower operating cost than Opus 5 and output more than 30% faster, while one report says the list cut is 20% and…
  • Terminal-Bench 4.0: Opus 5.5 66.4% versus Fable 5.1 55.8% and GPT-6 Astra 57.9%; other scores include FrontierCode v1.1 54.4%, FrontierCode 1.1 Extended 65.3%, CursorBench 57.8%, and OSWorld 2.0 81.8%. GPT-6 Astra still leads…
  • Cyber and biology safeguards are described as Fable-class. The Verge says cybersecurity routes to Opus 4.8 and flagged biology to Opus 5; Hacker News says Opus 4.8 or Opus 5; The Decoder also cites frontier-development routing.
  • Thinking cannot be disabled, outputs carry EU AI Act watermarking, and METR and Frontier Design tested Opus 5.5 before release. Anthropic claims better prompt-injection resistance than Opus 5 and its best behavioral-audit score.
  • Opus 5.5 is the default in Claude Code and the Claude app at medium effort. A 28 September guide says medium effort can match or beat Opus 5 at high effort and that thinking counts toward max_tokens, up to 128,000.
  • Sonnet 5.5, released 28 September, stays at $2/$10 per million tokens, is over 30% faster and up to 30% cheaper per task, scores 70.6% on Terminal-Bench 4.0 versus 10.3% for Sonnet 5, and is the first Sonnet shipped with cyber safeguards;…
  • OpenAI released GPT-6 Sol and Luna about an hour after Opus 5.5. Launch-day reports still listed Haiku 5.5 for the coming weeks.
Explain this AI news with Vani AIWhat was announced, what is new and why it matters, in a new chat
Vendors
AnthropicAmazon Web ServicesGoogleMicrosoftOpenAI
Products
Claude PlatformClaude CodeClaudeClaude Sonnet 5.5Claude Sonnet 5Claude Opus 5.5Claude API
Organizations
AnthropicDeloitteMETRFrontier DesignZapierOpenAIAmazon Web ServicesGoogle CloudMicrosoftGoogleEpic Games
AI models
Claude Opus 5.5Claude Fable 5.1Claude Opus 5GPT-6 AstraClaude Mythos 5.1GPT-5.6 SolClaude Opus 4.8Claude Sonnet 5.5Claude Haiku 5.5GPT-6 SolGPT-6 LunaClaude Fable

Coverage

oldest first · one row per article · UTC
  1. · Sep 22, 2026 Hacker News · AIFirst report

    Claude Opus 5.5

    Anthropic released Claude Opus 5.5, matching Fable 5.1 on most work at about 40% lower cost with stronger safeguards.

  2. · Sep 22, 2026 The Verge · AI

    Anthropic launches Claude Opus 5.5 with stricter safeguards for cybersecurity

    Anthropic launched Claude Opus 5.5 with cybersecurity and biology safeguards that reroute risky requests to weaker models.

  3. · Sep 22, 2026 TechCrunch · AI

    Anthropic releases Opus 5.5 with lower prices and Fable-level performance

    Anthropic released cheaper, faster Claude Opus 5.5 and called it its strongest model.

  4. · Sep 22, 2026 The Decoder

    Claude Opus 5.5 matches Fable 5.1 performance at lower cost and promises less "Claudish" writing

    Anthropic launched Claude Opus 5.5, matching Fable 5.1 at about 40 percent lower cost.