Claude Sonnet 5.5 is Anthropic's new mid-tier model, released September 28, 2026, and it landed second on the most-watched independent AI leaderboard. That's rare for a model that costs half as much per token as the flagship it sits beneath. On the Artificial Analysis Intelligence Index, Sonnet 5.5 scores 56, just two points behind Claude Opus 5.5 and comfortably ahead of OpenAI's competing models.

If you use AI assistants for everyday work, Sonnet 5.5 is the model you're most likely to actually touch, since mid-tier models power most free and cheap tiers. This explainer covers what the model is, how it scores, what it costs, and the catch most coverage skips.

What Claude Sonnet 5.5 Is

Anthropic names its models in tiers. Opus is the flagship, built for complex work requiring careful judgment. Sonnet is the middle child, billed in Anthropic's docs as "the best combination of speed and intelligence." Haiku is the fast, cheap option. Sonnet 5.5 is the second member of the Claude 5.5 family, following Opus 5.5, which launched September 22.

The positioning is modest on purpose. Anthropic says Sonnet 5.5 is strongest at well-scoped everyday tasks, at fixing bugs, and at creating polished documents, slides, and spreadsheets. Opus 5.5 is for the open-ended, judgment-heavy work. Even where the benchmarks are close, Anthropic notes Opus 5.5 "remains clearly stronger at complex, open-ended work requiring sustained judgment."

That's a useful frame. Sonnet 5.5 isn't trying to dethrone Opus. It's trying to be the model people reach for daily because it's fast and affordable, and the benchmarks suggest it does that job well.

Claude Sonnet 5.5 Benchmarks

The numbers tell a consistent story across two sources: Anthropic's own testing and independent verification.

Artificial Analysis puts Sonnet 5.5 at 56 on its Intelligence Index, ranking it third of 216 models, behind only Opus 5.5 at 58. The index combines ten evaluations, including Humanity's Last Exam, SciCode, and Terminal-Bench 4.0. With max effort, Sonnet 5.5 gains 18 points over Sonnet 5, a substantial generational jump.

On specific benchmarks, the figures hold up. Sonnet 5.5 scores 70.6% on Terminal-Bench 4.0, 81.3% on SWE-Bench Pro, 55.5% on CursorBench 4.0, 80.1% on OSWorld 2.1, and 1,844 Elo on GDPval-AA v2.1. That GDPval figure is two Elo points behind Opus 5.5, and about 400 ahead of Sonnet 5.

Here's the detail worth flagging: in Artificial Analysis's own runs, Sonnet 5.5 actually beat Opus 5.5 on Terminal-Bench 4.0, 64% to 60%. A cheaper model edging out the flagship on a coding benchmark is the kind of result that makes mid-tier releases interesting.

Sonnet 5.5 vs the Competition

Against OpenAI, Anthropic's own charts put Sonnet 5.5 ahead on some measures, including GDPval-AA and AA-Briefcase, while rival models edge it on others like FrontierCode's coding benchmark. The honest read is that the top AI models have clustered close together, and Sonnet 5.5 puts Anthropic's mid-tier squarely in that cluster rather than trailing it.

That clustering is the real story of 2026. The gaps between leading models have narrowed to a few points, which means picking a model is less about raw capability and more about cost, speed, and how each one fits your workflow. Sonnet 5.5's pitch is that it lands near the top on quality while staying cheap, and on the Intelligence Index it does exactly that.

Within Anthropic's lineup, the pricing tells the clearest story. Opus 5.5 costs $4 per million input tokens and $20 per million output. Sonnet 5.5 costs $2 and $10, exactly half. Cache reads are $0.20 per million on both. Haiku 4.5 sits below at $1 and $5 with a 200K context window, and Anthropic says Haiku 5.5 will join the family "in the coming weeks."

Anthropic also made an unannounced change here: Sonnet 5's $2/$10 price, originally labeled introductory, is now permanent, and a scheduled increase to $3/$15 was cancelled. Sonnet 5 becomes a legacy model, staying available but superseded.

What's New Beyond the Numbers

Sonnet 5.5 comes with a 1M-token context window, which is roughly 555,000 words, and 128K max output, expandable to 300K through a Batch API beta. It takes text and image input, with a knowledge cutoff of June 2026.

Anthropic says Sonnet 5.5 generates output more than 30% faster than Sonnet 5 and costs up to 30% less per task. That second claim deserves scrutiny, and it's where the catch lives.

Two smaller details matter for developers. Setting temperature, top_p, or top_k to a non-default value returns a 400 error, so tuning behaves differently than on older models. And fast mode, Anthropic's premium faster-output option, isn't offered on Sonnet 5.5 at all, since the pricing page lists it only for Opus models.

The Catch: Cost per Task

Sonnet 5.5 is cheap per token and can be expensive per task. That sounds like a contradiction, and it isn't.

Anthropic's efficiency claims about 30% lower cost per task apply at lower effort levels. Run Sonnet 5.5 at its highest effort setting and it burns so many tokens that independent testing puts it above Opus 5.5 on cost per task. Artificial Analysis measured $7.60 per task and 193K output tokens per task at max effort, compared with $5.98 for Opus 5.5 and $1.06 for GPT-6 Sol.

Those are wide gaps, and they reframe the whole decision. GPT-6 Sol has the same list price per token as Sonnet 5.5, but at max effort Sonnet 5.5 costs roughly seven times more per task. The lesson buried in the benchmarks is that the effort setting you choose matters more than the model you pick. Sonnet 5.5 makes sense as a fast, cheap work partner at moderate effort, not as a model you leave cranked to maximum and forget.

Claude Sonnet 5.5 What It Means for You

If you use an AI assistant tier built on Sonnet, you're getting a model that's close to flagship quality for routine tasks at half the token price. For writing, bug fixing, and document work, that's a strong deal. If your workflow leans on open-ended, high-judgment tasks, Opus 5.5 remains the better fit, and Anthropic says so itself.

The practical advice is to match effort to the job. Run Sonnet 5.5 at moderate effort for everyday work and reserve the higher settings for tasks that genuinely need them, because maxing it out erases the price advantage that makes the model attractive in the first place.

There's a data angle worth knowing if you use this model through a consumer app rather than the API. Models at this tier power many assistant experiences, and what you type into them is handled under the app's privacy policy, not yours. Don't paste credentials, sensitive records, or anything confidential into a chat prompt, and check the data controls of whichever app sits on top of the model. If you use a Claude app on Android, download it from a trusted store and keep Google Play Protect enabled, since a chat app that touches your account and history is worth installing carefully.

The Bottom Line on Claude Sonnet 5.5

Claude Sonnet 5.5 is the mid-tier model that finally makes "close to the flagship" a fair description rather than marketing. Its benchmarks back that up, its price is honest, and its main trap is a setting rather than a flaw. Use it where speed and cost matter most, keep an eye on effort, and it earns its place as a daily driver.

More ai tool: