TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Anthropic released Claude Sonnet 5.5 on September 28, 2026, claiming it runs 30%+ faster and costs up to 30% less for most work at the same price as Sonnet 5. Independent tester Simon Willison found it beats Sonnet 5 on benchmarks and now powers Claude’s free tier, which he judged more capable than ChatGPT’s free offering.
Anthropic released Claude Sonnet 5.5 on September 28, 2026, saying the new model “runs 30%+ faster, and costs up to 30% less for most work” than its predecessor while keeping the same price as Sonnet 5. According to independent testing by blogger Simon Willison, the model appears to beat Sonnet 5 on every benchmark — and it now also serves as the model behind claude.ai’s free tier, a move Willison says gives Anthropic a more capable free offering than ChatGPT’s.
The release was documented the same day by Simon Willison, the developer and blogger whose hands-on model evaluations are widely followed in the AI engineering community. According to Willison, Anthropic says Sonnet 5.5 runs 30%+ faster and costs up to 30% less for most work compared with Sonnet 5, despite being priced the same as the older model. Willison’s own benchmark runs suggested the new model appears to beat Sonnet 5 on every benchmark he tested, though these are informal evaluations rather than a formal third-party audit.
In Willison’s coding tests, Sonnet 5.5 came close to Anthropic’s larger Opus 5.5 on some tasks, including reproducing viral 3D animation tricks. That positioning matters for buyers: if a mid-priced model approaches the flagship’s coding performance at a fraction of the running cost, the flagship’s premium becomes harder to justify for routine work.
Not everything worked cleanly. Willison reported that Sonnet 5.5 suffers from the same bug he observed in Opus 5.5: at the highest “max” thinking-effort setting, the model spent 128,000 thinking tokens — costing $1.28 — before exhausting its token budget and failing to produce the requested SVG image. At the “xhigh” setting it completed the task in 41 seconds at a cost of 5.74 cents.
Why the Free Tier Upgrade Matters
The most consequential change, in Willison’s assessment, is that Sonnet 5.5 is now the model serving the free tier on claude.ai. He compared it directly with OpenAI’s free ChatGPT tier, which he says uses Luna 5.6, and concluded that Anthropic “currently have a much more capable free offering.” That is a single reviewer’s judgment, not a comprehensive evaluation, but if it holds up, it sharpens competition at the free tier — the entry point where most consumers first encounter these products.
For developers and businesses already paying for Sonnet 5, the economics improve without any price change: same listed price, faster responses, and lower effective cost for most workloads, according to Anthropic’s own claims. Willison’s testing supports the performance picture, though his evaluations are informal and based on his standard prompt suite, such as asking the model to draw a pelican riding a bicycle — a long-running stress test he uses to compare models.
The model tier below Sonnet remains in flux. Anthropic’s announcement reiterated that Haiku 5.5 will be available “in the coming weeks,” a gap that leaves Anthropic without a new small-model answer to competitors’ cheaper offerings in the meantime.
Anthropic’s Model Lineup Leading Up to 5.5
Anthropic’s Claude family is layered by capability and price: Opus at the top, Sonnet in the middle, and Haiku as the fast, low-cost option. Opus 5.5 was released earlier and set the current top-end benchmark; Willison’s testing found Sonnet 5.5 approaching it on coding tasks, continuing the pattern of mid-tier models closing the gap with flagships within months.
Willison’s pelican-on-a-bicycle test has become a recurring yardstick in his model reviews, exercising image generation, instruction-following, and cost-per-task. Under that test, Sonnet 5.5’s failure mode at maximum thinking effort mirrors one he documented in Opus 5.5, suggesting the behavior is shared across the 5.5 generation rather than specific to one model.
The competitive backdrop includes OpenAI’s GPT-6 line, including the Luna 5.6 model Willison identified as powering ChatGPT’s free tier. Anthropic’s move to put its fresh mid-tier model into the free tier reads as a direct response to that competition, though Anthropic has not framed it that way in the material reviewed.
“Runs 30%+ faster, and costs up to 30% less for most work.”
— Anthropic, in its announcement (as quoted by Simon Willison)
Unverified Claims and the Thinking Bug
Anthropic’s speed and cost figures — “30%+ faster” and “up to 30% less for most work” — are the company’s own claims and are hedged with qualifiers like “most work.” Independent verification beyond Willison’s informal tests is not yet available, and it is unclear what workloads fall outside the “most work” category.
The “max” thinking-effort bug remains unresolved: at that setting the model burned 128,000 tokens ($1.28) and failed to produce output. Willison observed the same failure in Opus 5.5, but Anthropic has not publicly confirmed or explained the behavior, and there is no announced fix.
Willison’s judgment that Claude’s free tier is “much more capable” than ChatGPT’s is one reviewer’s opinion based on a single prompt test — a WebGL pelican page — rather than a systematic comparison. The exact capabilities and limits of the free tier (rate caps, usage restrictions) were not detailed in the material reviewed.
Haiku 5.5 and Free-Tier Rivalry Ahead
Anthropic says Haiku 5.5 will arrive “in the coming weeks,” completing the 5.5 generation at the low end. Willison explicitly hopes it will be price-competitive with GPT-6 Luna, pointing to continued price pressure in the small-model segment.
Watch for whether Anthropic addresses the max thinking-effort failure shared by Opus 5.5 and Sonnet 5.5, and whether OpenAI responds by upgrading ChatGPT’s free tier beyond Luna 5.6. Broader third-party benchmarking of Sonnet 5.5 — beyond Willison’s informal runs — should emerge in the days after release and will test the company’s speed and cost claims.
Key Questions
What is Claude Sonnet 5.5?
It is Anthropic’s newest mid-tier AI model, released September 28, 2026. The company says it runs 30%+ faster and costs up to 30% less for most work than Sonnet 5, at the same listed price.
How much does Claude Sonnet 5.5 cost?
It is priced the same as Sonnet 5, according to Simon Willison’s review. Anthropic says effective costs drop up to 30% for most workloads because the model is faster and more efficient. Exact per-token rates for edge cases were not detailed in the material reviewed.
Is Sonnet 5.5 available on Claude’s free tier?
Yes. According to Willison, Sonnet 5.5 is now the model serving the free tier on claude.ai. He tested it there with a 3D WebGL prompt and judged the free offering more capable than ChatGPT’s free tier, which he says uses OpenAI’s Luna 5.6.
Is Sonnet 5.5 better than Opus 5.5 for coding?
Willison found it “almost as good as Opus 5.5 on some coding tasks,” including viral 3D animation tricks. It is a mid-tier model, so Opus 5.5 likely retains an edge on harder tasks, but formal independent comparisons are not yet published.
Are there known problems with Sonnet 5.5?
Yes, one reported issue: at the “max” thinking-effort setting, the model spent 128,000 tokens ($1.28) and failed to produce output, the same bug Willison documented in Opus 5.5. Lower effort settings such as “xhigh” worked normally, completing a task in 41 seconds for 5.74 cents.
Source: rss
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
