Anthropic has launched Claude Sonnet 5.5, claiming it runs more than 30% faster than Sonnet 5 and can cost up to 30% less per task. The release pairs those efficiency claims with a striking jump on an agentic coding benchmark, making this more than a routine model-number shuffle.
Anthropic Watch analysis
What happened
Sonnet 5.5 is available across Anthropic’s platforms, including its API, Amazon Web Services, Google Cloud and Microsoft Azure. Anthropic positions it as a faster, cheaper complement to Opus 5.5, particularly for everyday coding, fixing bugs and creating documents, slides and spreadsheets. The company’s announcement and evaluation details set out the changes and the limits of its comparisons.
Anthropic says Sonnet 5.5 scores 70.6% on Terminal-Bench 4.0, compared with 10.3% for Sonnet 5. It says the new model’s list price is unchanged at $2 per million input tokens and $10 per million output tokens, but that it typically uses fewer tokens for a task. The company also says Sonnet 5.5 generates output more than 30% faster.
Our top picks
- A much higher Terminal-Bench score
Anthropic reports 70.6% for Sonnet 5.5, against 10.3% for its predecessor, on this command-line task evaluation. - Lower claimed cost per task
The list price stays the same, but Anthropic says the model’s efficiency can reduce task costs by up to 30%. - More speed for iteration
Anthropic says output generation is over 30% faster than Sonnet 5, a useful difference when developers are testing and revising. - Better results at lower effort settings
Anthropic says Sonnet 5.5 at Low or Medium effort beats Sonnet 5’s best score on several evaluations for about a tenth of the task cost. - A new kind of screenshot test
Anthropic says this is the first Sonnet model to beat Pokémon Red using only screenshots, a memorable test of sustained visual interaction.
Why it matters
For developers, the pitch is better coding performance without automatically paying more per token, alongside fewer tokens per task and faster output. That could make Sonnet 5.5 attractive for routine agent work, where repeated tool calls and small edits can quietly turn into a sizeable bill.
The figures are Anthropic’s own evaluations and comparisons, not a guarantee of the same results across every codebase or workflow. The company also says Opus 5.5 remains clearly stronger on complex, open-ended work requiring sustained judgement. That distinction matters: a model can be a better everyday workhorse without being the best choice for the knottiest job on the desk.
Our read
This looks like a meaningful upgrade for practical coding and knowledge work, especially if the claimed efficiency gains hold up in real use. Anthropic has supplied enough detail to make the case interesting, though the benchmark table is still the company’s case. Teams should compare the models on their own tasks and include total task cost, not just the per-token rate. The new model is available now; a migration is a choice, not a ceremonial duty.
What to watch
- Whether independent evaluations reproduce the reported coding and cost gains.
- How Sonnet 5.5 performs on real codebases and longer agent workflows.
- Whether lower-effort settings deliver the promised balance of cost, speed and quality.
- How the model’s cyber safeguards affect routine development and higher-risk security work.
Discussion spark: Would you switch a coding workflow to a model with a much higher reported benchmark score and lower claimed task costs before independent tests confirm the gains?
Sources and evidence
- Introducing Claude Sonnet 5.5 – Anthropic (28 September 2026, 18:01 UTC)
Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.