Claude Sonnet 5.5: faster, cheaper per task, nearly Opus 5.5 level

Anthropic unveiled Claude Sonnet 5.5 on Monday, just days after Opus 5.5 arrived less than a week earlier. The positioning is straightforward: Opus 5.5 for complex work that demands careful judgment, Sonnet 5.5 for everyday, well-scoped tasks. In other words, a model meant to be used often without blowing up the bill.

One caveat up front: the speed, cost and performance figures come from Anthropic and have not been independently verified. The exact price and detailed availability are not spelled out in the source either. Treat them as vendor-stated ballparks, not third-party measurements.

What Anthropic claims

Three points stand out from the announcement as reported by Gizmodo:

• Speed: Sonnet 5.5 generates outputs at least 30% faster than Sonnet 5.
• Cost: up to 30% cheaper per task at the same per-token price — so the saving comes from how many tokens a task consumes, not from a lower unit rate.
• Agentic coding: Anthropic says Sonnet 5.5 scores better than Sonnet 5 on an agentic coding evaluation. The details of that evaluation are not described in the source.

The logic behind the first two points holds together: if a model produces the same answer in fewer tokens and less time, cost per task falls even when the per-token price is unchanged. That is exactly the argument being made for repetitive work.

Writing, coding, analysis: where Sonnet 5.5 makes sense

Writing and structured documents

Anthropic presents Sonnet 5.5 as suited to producing polished documents, slides and spreadsheets. That is the kind of work where you go back and forth: outline, rewrite, format, check. A faster, cheaper-per-task model lowers the cost of those iterations, which is often the real expense — more than the first draft.

Coding and bug fixing

Bug fixing is explicitly among the cited uses. Combined with the reported gain on the agentic coding evaluation, it points to a model built for short loops: reproduce, fix, rerun, verify. For well-scoped tasks, fast iteration matters more than depth of reasoning.

Analysis and tightly framed tasks

Anthropic says the speed makes Sonnet 5.5 suitable for rapid iteration on less complex tasks. That is a useful hint: for exploratory analysis where you want to try ten different angles, a fast, inexpensive model beats a frontier model used once. For a delicate trade-off, Opus 5.5 remains the stated option.

What Sonnet 5.5 does not change

Anthropic states that Sonnet 5.5 does not push the capability frontier of its models. The direct consequence: no new safeguards are announced for this release. The company also says Sonnet 5.5 was the least likely of its internally tested models to actively look for vulnerabilities in its sandbox. That claim rests on Anthropic's internal testing and has not been verified externally.

The Claude 5.5 family and the competitive backdrop

Sonnet 5.5 is the second member of the Claude 5.5 family. The third, Haiku 5.5, will be available "in the coming weeks" according to the company — a vague formulation with no precise date. The previous week, OpenAI had also launched two models, GPT-6 Sol and Luna, described as lighter, cheaper versions of Astra. Both competitors are pushing in the same direction: cheaper mid-tier models rather than capability leaps.

The contrast with the surrounding discourse is notable. CEO Dario Amodei's call to slow the AI frontier has not produced a noticeable change in Silicon Valley, and the risks of uncontrolled AI were a major theme at the UN General Assembly since the previous week. The Washington summit between Trump and Xi Jinping does not appear to have advanced cooperation on mitigating those risks; according to the White House, Xi Jinping agreed to use Trump's preferred phrase, "superintelligence."

Quick comparison

Key takeaways

• Keep Opus 5.5 for delicate trade-offs and complex work; use Sonnet 5.5 for volume.
• The claimed cost saving comes from tokens per task, not a lower unit price: measure your own workloads before concluding.
• For iterative writing and well-scoped bug fixing, speed matters more than depth.
• Treat the numbers as Anthropic's claims: the exact price, the Haiku 5.5 date and the details of the coding evaluation are all unspecified.

This article is published by Roboto, a platform for generating texts, images, videos and voiceovers with AI.

Source

Anthropic Releases Its Second New AI Model in Less Than a Week