With Claude Haiku 5.5 on October 7, Anthropic finished rolling out the Claude 5.5 family — three models in 15 days, each cheaper per task than the model it replaces.
The three releases
| Model | Released | Price (in / out per 1M) | What changed |
|---|---|---|---|
| Claude Opus 5.5 | Sep 22 | $4 / $20 (was $5 / $25) | Fable 5.1-level on most work; 40% lower cost per task than Opus 5; output 30%+ faster |
| Claude Sonnet 5.5 | Sep 28 | $2 / $10 (unchanged) | Terminal-Bench 4.0 70.6% (from 10.3%); up to 30% less per task; 30%+ faster |
| Claude Haiku 5.5 | Oct 7 | $0.10 / $0.50 for prompts ≤100K (was $1 / $5) | About 75% cheaper than Haiku 4.5; first Haiku with adjustable effort |
All three have a 1M-token context window, up to 128K output tokens and adaptive thinking, and run on the Claude Platform, AWS, Google Cloud and Microsoft’s platforms.
Why it matters
- Price cuts, not price hikes. The previous generation raised the ceiling; this one lowers it. Opus 5.5’s cache reads fall to $0.20 per million (from $0.50), and on October 7 Anthropic also halved Sonnet 5.5’s cache reads to $0.10 — which matters because cache reads dominate agent bills.
- The tier ladder got steeper. Opus 5.5 ($4 / $20), Sonnet 5.5 ($2 / $10) and Haiku 5.5 ($0.10 / $0.50) span a 40x price range for input tokens, and Anthropic explicitly pitches them as an orchestrator-plus-subagents stack.
- Fable 5.1 is now the niche pick. Anthropic says Opus 5.5 performs at Fable 5.1’s level on most work for under half the price, and tells customers to start with Opus 5.5 — reaching for Fable only when evals at higher effort still fall short.
- A head-to-head with OpenAI. On Anthropic’s own table Opus 5.5 beats GPT-6 Astra on Terminal-Bench 4.0 (66.4% vs 57.9%) while Astra leads on AutomationBench and scientific research. Both numbers are vendor-reported; wait for independent runs.
Caveats
- Safeguards sit on top of the models: higher-risk cybersecurity work falls back to older models, and thinking can no longer be turned off on Opus 5.5.
- Haiku 5.5’s price jumps 5x for prompts over 100K tokens ($0.50 / $2.50).
- Benchmark figures differ slightly between Anthropic pages (Sonnet 5.5’s GDPval-AA appears as 1844 and 1840), so treat decimals cautiously.
For the wider model landscape, see the 2026 LLM guide.