Term
Claude Sonnet 5.5
Claude Sonnet 5.5 (September 28, 2026) is Anthropic's low-cost mid-tier model — a faster, cheaper companion to Opus 5.5, priced at $2/$10 per 1M tokens and leading on Terminal-Bench 4.0. Available via boostN.
Claude Sonnet 5.5 — explained in more detail
Anthropic released Claude Sonnet 5.5 on September 28, 2026, just under a week after the flagship Opus 5.5 (September 22). The Sonnet line is Anthropic’s mid-tier, sitting between the Opus flagships and the small Haiku models. Sonnet 5.5 is positioned as a faster, cheaper companion to Opus 5.5 — Anthropic itself calls it “the cheap one” internally. The list price stays at Sonnet level ($2/$10), while the model, per Anthropic, costs up to 30 percent less per task because it does the same work with fewer tokens, and runs more than 30 percent faster than Sonnet 5.
On the independent Artificial Analysis Intelligence Index, Sonnet 5.5 scores 56 points, behind Opus 5.5 (58) but ahead of GPT-6 Astra (53) and GPT-6.1 Sol (52). Within the Claude family Opus 5.5 remains the stronger model; Sonnet 5.5 narrows the gap considerably and even beats the flagship on one benchmark.
Key facts
- Release: September 28, 2026 (Anthropic), about a week after Opus 5.5.
- Pricing (per 1M tokens): $2 input, $10 output, $0.20 cache read, $2.50 cache write (5 min), $1/$5 in batch. No surcharge in the 1M-token window. Same list price as Sonnet 5.
- Context: 1M tokens input, up to 128,000 tokens output (up to 300,000 in batch with a beta header). Knowledge cutoff June 2026.
- Reasoning effort: low, medium, high, xhigh, max. API default
high,mediumin Claude Code and the apps. Thinking is on by default and can only be reduced via{"type":"between_tools"}up to efforthigh. - Access: Claude API (model ID
claude-sonnet-5-5), Google Cloud, Microsoft Foundry and AWS Bedrock (anthropic.claude-sonnet-5-5), OpenRouter and the Vercel AI Gateway (anthropic/claude-sonnet-5.5), GitHub Copilot (rolling out) and Claude Code via thesonnetalias.
Benchmarks (quick overview)
Anthropic’s headline numbers apply only at maximum reasoning effort (max) — at the cheaper default tiers the figures drop sharply (Terminal-Bench 4.0, for example, is 70.6 percent at max but 43.0 percent at the API default high and 28.8 percent at the Claude Code default medium).
- Terminal-Bench 4.0: 70.6 percent — the only benchmark where Sonnet 5.5 beats the flagship Opus 5.5 (66.4 percent).
- OSWorld 2.1 (computer-use): 80.1 percent (Opus 5.5: 81.8 percent).
- GDPval-AA v2.1 (Elo): 1844, practically level with Opus 5.5 (1846).
- Artificial Analysis Intelligence Index: 56 — behind Opus 5.5 (58), ahead of GPT-6 Astra (53) and GPT-6.1 Sol (52).
Example / Practical use
Sonnet 5.5 targets cost-sensitive agentic and coding workloads: where many model calls accumulate, cost per task matters more than the absolute peak score. Its Terminal-Bench result shows the model can even overtake the pricier Opus 5.5 in agentic coding — but only at maximum effort and the corresponding cost per attempt. For everyday use the effort tier is decisive: the default medium yields noticeably lower benchmark scores than the advertised numbers.
Claude Sonnet 5.5 is available directly via boostN and ready in the model picker. In-house comparison tests against Opus 5.5 and GPT-6.1 Sol are still running.
Distinction
Sonnet 5.5 is not the most intelligent model on the market — within the Claude family Opus 5.5 stays ahead, and on the Intelligence Index Sonnet 5.5’s 56 points trail Opus 5.5 (58). Its edge is value: Sonnet pricing ($2/$10) at performance close to Opus 5.5 and higher speed. Against GPT-6.1 Sol (same $2/$10 list price), Sonnet 5.5 leads by four points on the Intelligence Index and is more than twice as fast at around 142 tokens per second; in return Sol 6.1 is cheaper per index task. Where maximum quality matters, Opus 5.5 remains the pick.