Anthropic's Claude Sonnet 5.5 Is Out, Beats Opus 5.5 at Coding for Half the Price
Anthropic launched Claude Sonnet 5.5, a mid-tier model priced at $2 per million input tokens and $10 per million output tokens — unchanged from Sonnet 5 and half the per-token price of Opus 5.5. Anthropic reports Sonnet 5.5 outperformed Opus 5.5 on Terminal-Bench 4.0, and an independent tester, Artificial Analysis, corroborated the lead while flagging that Sonnet 5.5 consumes more tokens per task than competing models.
Why It Matters
The release shows a mid-tier model can surpass a vendor's flagship on some coding benchmarks while undercutting flagship pricing, a dynamic that could reshape customer choices for cost-sensitive and code-focused use cases. At the same time, higher token consumption at certain effort settings may offset the headline price advantage in real workloads.
Key Facts
- Product: Claude Sonnet 5.5
- Pricing: $2 per million input tokens; $10 per million output tokens
- Relative price: Half the per-token price of Opus 5.5
- Anthropic Terminal-Bench 4.0 score: 70.6% for Sonnet 5.5 vs 66.4% for Opus 5.5
- Artificial Analysis Terminal-Bench 4.0 score: 63.6% for Sonnet 5.5 vs 59.6% for Opus 5.5; 59.1% for GPT-6 Astra
Anthropic has unveiled Claude Sonnet 5.5, an update to its mid-tier Sonnet line. The company kept Sonnet’s existing price points at $2 per million input tokens and $10 per million output tokens, the same as Sonnet 5 and roughly half the per-token cost of Anthropic’s flagship Opus 5.5. Anthropic says Sonnet 5.5 runs more than 30% faster than its predecessor and emphasized strengths in well-scoped everyday tasks, bug fixes, and polished document and design work.
On coding benchmarks, Anthropic reported Sonnet 5.5 outperformed Opus 5.5 on Terminal-Bench 4.0, a test that measures an AI agent’s ability to complete complex professional tasks via autonomous command execution. Anthropic’s result showed Sonnet 5.5 at 70.6% task completion versus 66.4% for Opus 5.5; Sonnet 5 scored 10.3%. Independent tester Artificial Analysis ran its own evaluation and likewise placed Sonnet 5.5 ahead of Opus 5.5 — 63.6% to 59.6% — and noted Sonnet 5.5 also beat OpenAI’s GPT-6 Astra in their run.
Despite the benchmark lead and lower list price, Artificial Analysis flagged a downside: Sonnet 5.5 produced more output tokens per task than any model it tested. At the highest effort setting the tester recorded roughly 193,000 tokens per task for Sonnet 5.5, about 60% more than Opus 5.5, which raised the per-task cost to approximately $7.60 — about 50% higher than Sonnet 5 on the same tests. Anthropic’s cost-saving claims focus on lower effort settings (Medium effort being the default in its apps), where it says Sonnet 5.5 can beat Sonnet 5’s best coding score for less than a tenth of the cost; Artificial Analysis judged High effort provided the best value in its runs.
On broader occupational skill evaluation (GDPval-AA across 44 occupations using Elo-style scoring), Sonnet 5.5 and Opus 5.5 were effectively tied, with scores of 1844 and 1846 respectively; GPT-6 Sol scored 1487. Anthropic described Opus 5.5 as remaining stronger on complex work requiring sustained judgment. Competing vendors have moved prices as well: OpenAI recently trimmed GPT-6 Sol to $2/$10, and GPT-5.6 Terra lists at $2/$12. Anthropic also said a high-volume, low-cost model called Claude Haiku 5.5 is planned to ship in the coming weeks.
Keep Reading
After AI Agent Hacked Its Government, Australia Calls Altman and Amodei to Testify
A Clever RSA Attack Fooled a Hardware Vault—Here's What It Means for Crypto
