Anthropic: Claude Sonnet 5.5 (batch)

Claude Sonnet 5.5 (batch) is the same Claude Sonnet 5.5 frontier model served through Anthropic's Message Batches API, which processes large asynchronous workloads at roughly half the standard input and output token cost. It targets high-volume agentic coding, knowledge work, and tool-use pipelines where latency is less critical than throughput and unit economics. The batch endpoint supports the full Sonnet 5.5 capability set, including adaptive extended thinking, multimodal image inputs, and a native one-million-token context window, with an optional beta path for up to 300k output tokens per request on batches.

Relative to Claude Sonnet 5, Sonnet 5.5 is a major step up on real-world software engineering and terminal agent benchmarks while running about 30% faster and often using fewer tokens for the same tasks. Anthropic's September 2026 system card reports strong scores on SWE-Bench Pro (81.3%), SWE-Bench Multilingual (90.3%), DeepSWE v1.1 (71.0%), GDPval-AA v2.1 (1844 Elo), Humanity's Last Exam with tools (64.5%), and Toolathlon-Verified Pass@1 (77.8%) at max adaptive thinking effort. Independent evaluations on MMLU-Pro and GPQA Diamond reach about 91% and 95.6% respectively under standardized harnesses.

Sonnet 5.5 is positioned as the fast, cost-efficient tier in the Claude 5.5 family: near Opus 5.5 on many professional and computer-use tasks but at lower per-token pricing, making the batch variant especially attractive for offline evaluation sweeps, dataset labeling, and large-scale code or research agents. Knowledge and training cutoffs are June 2026; the model was released on 2026-09-28 and is available on the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry under the ID claude-sonnet-5-5.