Anthropic released Claude Sonnet 5.5 at the same per-token price as Sonnet 5. It says the new model generates output more than 30% faster and can finish many jobs in fewer steps. Its reported coding jump is substantial: on Anthropic's Terminal-Bench 4.0 run, completed tasks rose from 10.3% for Sonnet 5 to 70.6% for Sonnet 5.5.
The more useful question is what you pay for a finished job. Artificial Analysis independently tested the model and found that Sonnet 5.5 approaches Opus 5.5 at maximum effort. At that setting it also uses more output tokens and costs more per task than Sonnet 5. A flat token rate and a lower total bill are not the same promise.
A coding leap, without a blanket Opus replacement
Terminal-Bench asks an agent to work through technical tasks in a command-line environment. Anthropic reports 70.6% for Sonnet 5.5 against 10.3% for Sonnet 5. Artificial Analysis ran its own configuration and measured 64% for the newer model. The numbers describe different runs, so the exact scores should not be spliced into one league table. Both point to a major improvement over the previous Sonnet.
On Artificial Analysis's broader Intelligence Index, Sonnet 5.5 reaches 56 points at maximum effort, just two behind Opus 5.5. Anthropic still says Opus is stronger for open-ended work requiring sustained judgment. A bounded bug fix or a first draft is where Sonnet's speed may be especially attractive. An ambiguous architecture decision can still justify Opus 5.5.
The same API price can buy a different-sized bill
Anthropic charges $2 per million input tokens and $10 per million output tokens for Sonnet 5.5, matching Sonnet 5. Cached reads cost $0.20 per million tokens. Tokens are pieces of material the model reads and writes. If one version writes substantially more while solving a task, an unchanged rate does not hold the overall cost steady.
Anthropic says Sonnet 5.5 costs up to 30% less per task in much of its testing. At maximum effort in Artificial Analysis's index, however, it produced roughly 193,000 output tokens per task and averaged $7.60, about 50% more than Sonnet 5 in that comparison. Higher effort can buy higher quality at a higher total cost. That is the number to test in your own workflow, alongside completion time. GPT-6 Sol shares the headline $2/$10 input-output rate, but price parity alone cannot tell you which finishes a task more cheaply.
Where it is available
Claude Sonnet 5.5 is available across Claude's platforms and through AWS, Google Cloud and Microsoft Azure. Developers can call `claude-sonnet-5-5` on Anthropic's platform. The company is positioning it for everyday coding, documents, slides, spreadsheets and visual design. We have not run our own head-to-head tests on those tasks.
Anthropic has also added safeguards for higher-risk cybersecurity requests, which can visibly fall back to Sonnet 5. That matters when interpreting an apparent Sonnet 5.5 result. The release looks strong, particularly for well-scoped work, but a sound choice still depends on the completed task, response time and total spend rather than a single chart.
The conversation starts here
Sign in with a supporter account to comment. Sign in
Nobody has commented yet. Want to go first?