Claude Opus 4.6

Anthropic announced Claude Opus 4.6 on February 5, 2026, upgrading its flagship model with stronger coding, agentic, and reasoning capabilities. The release introduces a 1M token context window in beta—the first for an Opus-class model—enabling operation across larger codebases and longer sustained agentic tasks. Claude Code also gains agent team assembly for collaborative task work. On benchmarks, Opus 4.6 achieves the highest score on Terminal-Bench 2.0 for agentic coding, leads on Humanity's Last Exam for multidisciplinary reasoning, and outperforms GPT-5.2 by approximately 144 Elo on GDPval-AA economically valuable knowledge work tasks. The API adds adaptive thinking, new effort controls, and a compaction capability that lets Claude summarize its own context for longer-running tasks. Safety evaluation rates are described as comparable to or better than other frontier models.

Anthropic announced Claude Opus 4.6 on February 5, 2026, introducing multi-agent teams in Claude Code (research preview) for orchestrating autonomous subagents on parallel tasks like code reviews and debugging. The launch also debuted a beta 1M-token input context window on the Developer Platform, paired with context compaction for coherence over prolonged interactions, and supports up to 128K output tokens. The model shipped at unchanged pricing of $5/$25 per million input/output tokens and added effort-level controls (low to max) alongside adaptive thinking for cost and intelligence tuning. Reported benchmark highlights include top performance on Terminal-Bench 2.0, 76% on MRCR v2 1M variant, and the highest score on Humanity's Last Exam, positioning Opus 4.6 as an agentic coding and long-context leader at launch.

CNBC reported on February 5, 2026 that Anthropic launched Claude Opus 4.6, describing it as the company's first major model release of 2026, following the Opus 4.5, Sonnet 4.5, and Haiku 4.5 family launched late the previous year. The report highlighted improvements in coding capability, longer task sustainment, and higher-quality professional output. Enterprise customers make up roughly 80% of Anthropic's business, CEO Dario Amodei told CNBC, and the company's Claude Code tool and Claude Cowork productivity product have drawn attention from software-sector investors. Anthropic head of product for enterprise Scott White framed the release as a transition toward "vibe working," extending the earlier "vibe coding" concept to broader professional workflows.

Anthropic's official documentation confirms Claude Opus 4.6 carries the model ID `claude-opus-4-6` and is available across Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and Claude Platform on AWS. It offers a 1M-token context window, 128K max output (300K via Batch beta), adaptive thinking, and is priced at $5 input / $25 output per million tokens. The model was released February 5, 2026, with a May 2025 knowledge cutoff and August 2025 training cutoff. Per the same documentation, Opus 4.6 now holds "Active (legacy)" status, superseded by Claude Opus 5.5 and Claude Fable 5.1 in the current lineup, with retirement not sooner than February 5, 2027. The docs provide resources for migrating to Opus 5.5, indicating Opus 4.6 remains supported but is no longer the flagship. This status update is the most current authoritative information on the model's lifecycle position.

Claude Opus 4.6 scores 64.3/100 on the benchlm.ai composite capability assessment (31st of 196 ranked models), with pricing at $5 input / $25 output per million tokens and a 1M token context window. The model achieves 39 tokens per second output speed with a 2.23-second first-token latency. It has been superseded by ne Category-level performance shows Agentic at 44.2 (rank 40 of 105, 63rd percentile), Coding at 51.4 (rank 37 of 135, 73rd percentile), Multimodal at 59.6 (rank 30 of 50, 41st percentile), and Knowledge at 57.3 (rank 36 of 160, 78th percentile). Agentic is its lowest-scoring eligible category at 40. The data is current a

A research summary on Emergent Mind, updated March 12, 2026, evaluates Claude Opus 4.6 as a frontier LLM for enterprise and security applications using rigorous benchmark criteria from the Corecraft and TEE-RedBench studies. The Corecraft-Expanded environment within EnterpriseGym reports task pass rates of 22.10% for C The analysis notes that Claude Opus 4.6 demonstrates advanced reasoning and tool-use capabilities but is constrained by complex multi-step demands in novel enterprise and security workflows, with notable failure modes including poor search strategies, incomplete tool exploration, and challenges in multi-step planning.

A preprint published on Preprints.org on February 6, 2026 presents a comprehensive comparative analysis of Claude Opus 4.6 alongside OpenAI's GPT-5 series, Google's Gemini 2.5/3 Pro, and GLM-4.6. The paper highlights Claude Opus 4.6's introduction of a 1 million token context window as a significant milestone and ident Performance evaluation spans automated coding, medical informatics, regulatory document processing, and general reasoning benchmarks, positioning Claude Opus 4.6 as a leading model in the early-2026 LLM landscape. The preprint is authored by Satyadhar Joshi and is not peer-reviewed. The paper explicitly names Claude Op