Claude Sonnet 5.5

Anthropic released Claude Sonnet 5.5 on September 28, 2026, positioning it as the second model in the Claude 5.5 family and targeting well-defined everyday work like bug fixing, documentation, presentations, and spreadsheets. It generates output more than 30 percent faster than Sonnet 5 and cuts per-task costs by up to 30 percent through more efficient token usage, rather than a sticker-price cut. On Terminal-Bench 4.0, Sonnet 5.5 hits 70.6 percent versus Sonnet 5's 10.3 percent, nearly matching Opus 5.5. Anthropic also added new safeguards against cybersecurity risks and distillation attacks for the more capable model. Claude Haiku 5.5 was announced for coming weeks, focused on high-throughput, low-cost use cases. The model became available on AWS, Google Cloud, and Azure at launch. Sonnet 5.5 sits between Opus 5.5 (complex tasks demanding careful judgment) and the upcoming Haiku 5.5 in Anthropic's refreshed lineup.

Vellum detailed Anthropic's Sonnet 5.5 benchmarks released September 28, 2026. On Terminal-Bench 4.0 for command-line agentic coding, Sonnet 5.5 scored 70.6 percent at peak effort, jumping from Sonnet 5's 10.3 percent and edging past Opus 5.5's 66.4 percent at extra-high effort. On GDPval-AA v2.1 for professional knowledge work, it scored 1844 Elo, two points shy of Opus 5.5 and leaving OpenAI's GPT-6 Sol nearly 360 points behind. List pricing holds at $2 input, $10 output, and $0.20 per million prompt cache read tokens. Sonnet 5.5 reduces effective cost per task by up to 30 percent by resolving problems in fewer steps and batching tool executions more aggressively. At medium effort, Sonnet 5.5 exceeds Sonnet 5's best Terminal-Bench score for less than a tenth of the per-task cost by avoiding unnecessary subagent spawning.

VentureBeat confirmed Anthropic's September 28, 2026 launch of Claude Sonnet 5.5 with API pricing held at $2 per million input and $10 per million output tokens, plus $0.20 per million cache-read tokens and $2.50 per million cache writes. The 30 percent cost reduction comes from fewer tokens and tool calls per task rather than lower API rates. Opus 5.5 pricing for comparison is $4 input and $20 output per million tokens. On GDPval-AA measuring real-world occupational tasks, Sonnet 5.5 scored 1844 versus Opus 5.5's 1846 and Sonnet 5's 1449. On AA-Briefcase, Sonnet 5.5 reached 1811 against Opus 5.5's 1822. The model also hits 80.1 percent on OSWorld 2.1, Anthropic's computer-use benchmark. Anthropic pitches the model for debugging, coding, documents, presentations, and interface refinement work.

BenchLM's aggregator card lists Claude Sonnet 5.5 with a capability score of 80.5 out of 100 across 209 ranked models, with coding ranked 2 of 142 (99th percentile) and knowledge ranked 6 of 168 (97th percentile). The card confirms 1M token context, $2 input and $10 output pricing, $0.20 cached reads, and $0.10 batch with cache rates. Agentic capability scored 66.2 at rank 10 of 117. The benchmark card shows 47 published benchmark rows with verified evidence state, though independent runtime speed was not measured. Multimodal capability scored 83.3 across 7 verified benchmarks. Coding category weights 20 percent with 9 verified benchmarks, making it particularly well-suited for software development and code generation tasks according to the aggregation.

CodeRabbit ran Sonnet 5.5 through its review pipeline on September 28, 2026, finding it caught 6 of 13 known-bug cases through actionable comments versus 4 for Sonnet 5, at 41.2 percent actionable precision. The model delivered results in about half the wall-clock time, with Claude model calls costing about 60 percent less per review, roughly twice the savings Anthropic advertises. A larger run on 44 open-source pull requests confirmed the speed and comment-volume improvements, marking the first Sonnet release CodeRabbit would consider for the main review pass rather than only for comment quality. Four bugs caught were ones Sonnet 5 waved through; Sonnet 5.5 missed two that Sonnet 5 caught, indicating different misses as much as more catches overall.

Vals AI's independent evaluation lists Claude Sonnet 5.5 with a September 28, 2026 release date, 1M token context window, 128k max output, and $2.00/$10.00 per million input/output token pricing. The model achieved a Vals Index of 69.22 percent plus or minus 0.96, ranking second of 66 models tested, with Code Migration at 69.83 percent, EMB at 75.71 percent, and Vibe Code Bench at 92.39 percent. Other notable Vals AI scores include Terminal-Bench 4.0 at 53.03 percent, ProofBench v1.1 at 100 percent, MedScribe at 91.10 percent, and IOI at 83.06 percent. Latency registered at 70 minutes 7 seconds with a 2.27 percent fallback rate and 0.14 percent refusal rate. The default provider listed is Anthropic, with text and image input supported but video not supported.

DataCamp framed Sonnet 5.5 as Anthropic's mid-tier model replacing Sonnet 5 at the same price, arriving six days after Opus 5.5 and closing most of the capability gap at half the per-token price. The publication noted this pattern of mid-tier models matching flagship benchmarks at lower cost reflects a logarithmic scaling curve, where most value is cheap and only the final stretch is expensive. Anthropic's own documentation describes Sonnet 5.5 as the best combination of speed and intelligence in the current lineup, sitting between Opus 5.5 and the upcoming Haiku 5.5. The headline gain is agentic terminal coding, where it edges past Opus 5.5. On knowledge work and computer use, it trails Opus 5.5 by only a couple of points for half the price.