Anthropic’s Claude Sonnet 5.5 almost equals Opus 5.5 in benchmark performance while being up to 30% cheaper per task

Meta description: Anthropic Claude Sonnet 5.5 timeline: release, benchmark results, pricing changes, cybersecurity safeguards, and what it means for everyday AI work.

Tags: Claude Sonnet 5.5, Anthropic, AI models, benchmark, pricing

Featured image: Abstract blue network nodes and data paths on a clean dark background, no text, no logos, no people.

Background: Anthropic Expands the Claude 5.5 Family

Anthropic has released Claude Sonnet 5.5, the second model in its Claude 5.5 family. It produces output at more than 30 percent higher speed, lowers cost per task by up to 30 percent, and is close to Opus 5.5 on multiple benchmark tests. Anthropic frames it as a model for well-defined everyday work, including bug fixes, documentation, presentation creation, and spreadsheet building. Opus 5.5 remains aimed at difficult work where judgment matters most. Claude Haiku 5.5 is expected in the coming weeks for high-throughput, low-cost use cases.

The release adds Claude Sonnet 5.5 to a broader line-up (Fable, Opus, and Sonnet) that already maps onto OpenAI's GPT-6 Astra, Sol, and Luna, the models that started the latest pricing battle. In rough tier terms, Opus sits a bit above Sol, Sonnet above Luna, and Fable above Astra. Anthropic charges more across the board, while matched-tier performance gaps are small enough that cost may become the deciding factor. Haiku may help close that gap.

Development: Benchmarks, Pricing, and Capability Gaps

Coding is where the jump is largest. On Terminal-Bench 4.0, a test for agentic coding, Sonnet 5.5 reaches 70.6 percent, compared with 10.3 percent for Sonnet 5. On CursorBench 4.0, which recreates real coding sessions from the Cursor editor, it scores 55.5 percent versus 34.1 percent, just two points below Opus 5.5 at 57.8 percent. On FrontierCode 1.1 at the High setting, Sonnet 5.5 scores ten points above Sonnet 5 at roughly one-fifteenth the cost per task. Anthropic says early testers praised how quickly it grasps a codebase, and that it batches tool calls more often, reducing steps.

A reasoning wrinkle appears at the highest effort level, Max: Sonnet 5.5 scores worse on FrontierCode than at Xhigh. Anthropic attributes this to a code-review function that more often splits work across multiple sub-agents, sometimes causing timeouts or out-of-scope changes that the benchmark penalizes. Anthropic also notes that Opus 5.5's highest Terminal-Bench 4.0 score comes at Xhigh, and that Sonnet 5.5 values come from a pre-release version where a since-fixed bug may have affected structured outputs.

On knowledge work, GDPval-AA, an OpenAI-developed benchmark covering tasks from 44 professions and nine industries, gives Sonnet 5.5 1,844 points, nearly matching Opus 5.5 at 1,846 and about 400 points above Sonnet 5 at 1,449. Anthropic places OpenAI's GPT-6 Sol at 1,487 by Anthropic's numbers. On Chartography, a visual chart recognition test, Sonnet 5.5 jumps from 15.6 to 61.6 percent. Early testers described it as a more natural conversational partner with a feel for design, and Anthropic says it can rework user interfaces and execute slide templates so well that results barely need touch-ups. The company also says it is the first Sonnet model that can play through Pokémon Red using only screenshots. OpenAI's Astra recently showed similar progress on gaming benchmarks, underscoring how tasks once considered hard and requiring custom algorithms are now handled by mid-tier models.

Pricing stays at Sonnet 5 levels per million tokens: $2 for input tokens, $10 for output tokens, and $0.20 for cache reads. Because tasks consume fewer tokens, effective cost drops by up to 30 percent, and output generation is more than 30 percent faster. Independent testing still needs to confirm these claims. Like other Anthropic and OpenAI models, Claude Sonnet 5.5 offers an adjustable effort setting that lets users trade cost and speed against output quality. At low or medium effort, it can already beat Sonnet 5's best scores on several benchmarks at about one-tenth the per-task cost, though choosing the right level remains more art than science.

Latest Progress: Availability, Cybersecurity Guardrails, and What to Watch

Claude Sonnet 5.5 is available now across all major cloud platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure. Like Opus 5.5 and Sonnet 5, it can be used with zero data retention, and developers can call it from the Claude Platform through the model ID "claude-sonnet-5-5".

Because the model is far more capable in cybersecurity than its predecessor, Anthropic is adding safeguards to a Sonnet model for the first time. Requests involving high-risk cybersecurity tasks are visibly rerouted to Sonnet 5, while qualified professionals can apply for tiered access through an expanded Cyber Verification Program. The company has also added safety classifiers against distillation attacks, the same kind used in its most powerful models. A further question is whether these measures work; if Chinese labs have been benefiting from such attacks, closing that vector could widen the gap with Western labs again.

For now, the timeline points to a mid-tier model approaching flagship workloads: Claude Sonnet 5.5 is deployed, Haiku 5.5 is pending, and external evaluations will determine whether the benchmark gains translate into durable cost and capability advantages.