AI developer tools · What shipped

Anthropic released Opus 5.5 on Tuesday, a new state-of-the-art model that outperforms the larger Fable model across many benchmarks and succeeds in tasks where Fable failed. The model features significantly lower pricing at $20 per million output tokens (down from $25) and improved speed due to reduced compute requirements. Opus 5.5 also includes refined communication patterns with less jargon and better information prioritization. Sonnet 5.5 and Haiku 5.5 will follow in the coming weeks with similar performance gains. The release reflects Anthropic's deliberate pacing strategy announced by CEO Dario Amodei to balance capabilities advancement with alignment risk prevention, and includes comparable safety training and pre-release evaluation by organizations like METR and Frontier Design.

Anthropic released Claude Fable 5.1 and Claude Mythos 5.1, with Fable 5.1 representing significant performance improvements across multiple benchmarks. Fable 5.1 achieves 52.6% on Terminal-Bench-Science 0.1 (agentic scientific research), 55.8% on Terminal-Bench 4.0 (agentic coding), 60.9% on Humanity's Last Exam (multidisciplinary reasoning), and 73.4% on CursorBench 3.2.0 (agentic coding)—all improvements over Fable 5. Pricing has been reduced by approximately 25% for typical workloads and up to 45% for highly agentic work through 75% cheaper cache reads at $0.25 per million tokens, while maintaining base rates of $10 per million input tokens and $50 per million output tokens. Claude Mythos 5.1, identical to Fable 5.1 but with reduced safeguards, demonstrates advanced scientific capabilities including protein design with hit rates reaching nearly 50% (versus typical 10-15%), GPU kernel optimization achieving up to 2.5x speedup on deep learning models, and creation of a high-resolution Venus elevation map at 300m resolution. The models include improved safeguards with 85% fewer false positives for benign biology requests and 60% fewer cybersecurity intervention rates, while introducing anti-distillation mechanisms and Enterprise Frontier Safeguards for zero-data-retention deployment.

Anthropic released Claude Opus 5.5 in September 2026, a proprietary model family consisting of five variants with adaptive reasoning capabilities. The models span intelligence scores from 42 to 58 on Artificial Analysis' Intelligence Index v4.3.2, output speeds from 76 to 90 tokens per second, and pricing from $0.55 to $5.98 per task across different effort levels (low, medium, high, xhigh, max). The benchmark incorporates 10 evaluations including AA-Briefcase, GDPval-AA, AutomationBench-AA, Terminal-Bench, SciCode, Humanity's Last Exam, and others, with all variants supporting a 1M token context window, text and image input, and text output.