Anthropic ships Sonnet 5.5 — its second new model in under a week
Bottom line (updated November 10, 2026): Claude Sonnet 5.5 is Anthropic’s value flagship — near-Opus capability for agentic coding and everyday professional work at exactly half Opus 5.5’s API price. If you run well-scoped tasks at volume, this is the Claude tier to default to; reserve Opus for the hardest reasoning.
Published September 30, 2026. Anthropic is releasing frontier models at a blistering pace. Just days after debuting Opus 5.5, the company has launched Claude Sonnet 5.5, a faster, cheaper mid-tier model — even as CEO Dario Amodei publicly calls for the industry to slow down and let safety frameworks catch up.
What’s new
- 30% faster outputs than Sonnet 5, with significantly slower token burn — meaning lower operating costs for high-volume use.
- Positioned for “well-scoped everyday tasks”: fixing bugs, drafting polished documents, slides, and spreadsheets.
- Anthropic’s benchmarks show Sonnet 5.5 outperforming Opus 5.5 on agentic coding tasks — likely because it can spin up multiple parallel agents without blowing cost limits.
- It’s the first Sonnet-class model with cyber capabilities comparable to Opus 5, so it now carries the same cyber safeguards as Anthropic’s larger models.
- Haiku 5.5, the smallest of the Claude 5.5 family, has since joined the lineup as the budget tier.
Verified pricing (checked November 10, 2026)
Anthropic’s own pricing lists Claude Sonnet 5.5 (model ID claude-sonnet-5-5) at $2 per million input tokens and $10 per million output tokens, with a 1M-token context window — exactly half the $4 / $20 charged for Opus 5.5. A detail most coverage missed: Anthropic cut Sonnet 5.5 cache reads to $0.10 per million tokens on October 7, so repeated long prompts are cheaper than the launch-day figures suggested. Haiku 5.5 lists from $0.10 / $0.50 for prompts up to 100K tokens. For how the tiers feel in daily use, see our full Claude review.
On agentic coding specifically, compare Sonnet 5.5’s pitch with the dedicated tools in our GitHub Copilot review and our Augment Code review before moving a whole workflow over.
Why it matters
The mid-tier is where the value war is being fought. If Sonnet 5.5 really does beat Opus 5.5 at agentic coding while costing less, the “just buy the biggest model” logic breaks down — good news for anyone watching their API bill. We compare it against GPT-6.1 Sol in our OpenAI DevDay coverage.
Sources: Gizmodo, Complete AI Training.