Anthropic releases Claude Sonnet 5.5, now the claude.ai free tier model
On 28 September 2026 Anthropic released a new Sonnet model, Claude Sonnet 5.5. A blog post about the release quotes Anthropic as saying the model "runs 30%+ faster, and costs up to 30% less for most work". The blog author adds that it is priced the same as Sonnet 5, appears to beat it on every benchmark, and should be cheaper to run as well.
The author tested it with the usual pelican-on-a-bicycle SVG prompt. Sonnet 5.5 suffered from the same bug as Opus 5.5: at the "max" thinking effort the pelican run thought for 128,000 tokens, at a cost of $1.28, then ran out of tokens and failed to produce an SVG. At the "xhigh" thinking effort the model did produce a pelican, at a cost of 5.74 cents and taking 41 seconds.
On coding, the author says Sonnet 5.5 appears to be almost as good as Opus 5.5 on some tasks, including various viral 3D animation tricks.
The author calls the most interesting part of the release the fact that Sonnet 5.5 is now the model used for the free tier on claude.ai. OpenAI's ChatGPT free tier uses Luna 5.6, which the author takes to mean Anthropic currently has a much more capable free offering. To try it, the author ran this prompt against the free tier: "build me an HTML page that renders a three-dimensional pelican riding a bicycle using WebGL". The page that came back was, in the author's words, a solid effort.
Finally, Anthropic's announcement reiterates that Haiku 5.5 will be available "in the coming weeks". The author hopes that model will be price-competitive with GPT-6 Luna.
Key facts
- Anthropic released Claude Sonnet 5.5 on 28 September 2026; it says the model "runs 30%+ faster, and costs up to 30% less for most work".
- Sonnet 5.5 is priced the same as Sonnet 5 and, per the blog author, appears to beat it on every benchmark.
- It is now the model behind the free tier on claude.ai; the author contrasts this with ChatGPT's free tier, which uses Luna 5.6.
- In the author's pelican test, the "max" thinking effort burned 128,000 tokens ($1.28) and produced no SVG, while "xhigh" delivered one in 41 seconds for 5.74 cents.
- Anthropic's announcement says Haiku 5.5 will arrive "in the coming weeks".
Why it matters
The author sees the free tier as the most interesting part. Sonnet 5.5 now runs the free tier on claude.ai, while OpenAI's ChatGPT free tier uses Luna 5.6, and the author concludes that Anthropic currently has a much more capable free offering. Beyond that, Anthropic pairs a same-price release with a claimed speed gain of 30%+ and a cost reduction of up to 30% for most work. The author also finds it almost as good as Opus 5.5 on some coding tasks.
Who it affects
Anyone using the free tier on claude.ai, which now runs on Sonnet 5.5. Developers and teams already on Sonnet 5 are affected too: the price is the same, and the claimed gains are in speed and cost. People waiting on a cheaper small model should note that Anthropic says Haiku 5.5 is coming in the coming weeks.
How to use it
The free tier on claude.ai already runs Sonnet 5.5, so trying it takes no setup. The author tested it there with a WebGL prompt for a three-dimensional pelican on a bicycle and called the result a solid effort. On price, the only detail given is that Sonnet 5.5 costs the same as Sonnet 5; no absolute prices for Sonnet 5.5 or Sonnet 5 (per million tokens) are given. In the author's pelican test the "xhigh" thinking effort worked and the "max" effort did not.
How solid is it
The speed and cost figures are Anthropic's own claim, quoted in the post; the source does not say what they are measured against or whether the author verified them. The claim that the model beats Sonnet 5 on every benchmark is the author's hedged impression ("appears to"); no benchmark names or scores are given. The hands-on evidence is informal: pelican SVGs at different thinking efforts and one WebGL prompt on the free tier. The statement that Sonnet 5.5 is almost as good as Opus 5.5 is likewise hedged and limited to some coding tasks.
Risks and caveats
Sonnet 5.5 suffered from the same bug as Opus 5.5: at the "max" thinking effort the pelican run used 128,000 tokens, cost $1.28 and failed to produce an SVG. The source does not say what the bug is, or whether Anthropic acknowledged or fixed it. The cost figures, $1.28 for the failed run against 5.74 cents for the working "xhigh" run, show how expensive a run that fails can be. Haiku 5.5 has no release date beyond "in the coming weeks" and no price.
“runs 30%+ faster, and costs up to 30% less for most work”
— Anthropic, as quoted in a blog post about Claude Sonnet 5.5