AI developer tools · What shipped

Anthropic released Claude Fable 5.1 and Claude Mythos 5.1, advanced models designed for coding and knowledge work. Fable 5.1 shows significant performance improvements across multiple coding benchmarks: Terminal-Bench 4.0 (55.8% vs Fable 5's 42.0%), CursorBench 3.2.0 (73.4% vs 70.5%), and Humanity's Last Exam (65.0% with tools vs Fable 5's 63.8%). Pricing is reduced approximately 25% for typical workloads and up to 45% for agentic tasks through 75% cheaper cache read pricing ($0.25 per million tokens). The model achieves improved inference efficiency: Mythos 5.1 optimized seven open-source genomics and protein models with speedups up to 2.5x on NVIDIA H100, reducing estimated GPU costs by 30–60% on genome-wide analyses. Fable 5.1 demonstrates enhanced code debugging capabilities—it identified a rare crash in production systems that engineering teams had failed to explain over several years. Safeguards were refined to reduce false positives by 60% in cybersecurity contexts while now enabling vulnerability discovery work. The models are available across AWS, Google Cloud, and Microsoft Azure platforms with API access at $10 per million input tokens and $50 per million output tokens.

Claude Fable 5 leads the September 2026 AI model leaderboard at 100/100 on a composite quality index across 56+ models ranked by quality, speed, pricing, and value using benchmarks including MMLU Pro, HumanEval, MATH, and LMSys Arena Elo ratings. DeepSeek V4 Flash offers the cheapest inference at $0.14/$0.28 per 1M tokens with cache hits at $0.0028, while open-weight models like Moonshot's Kimi K3 (ranking #3 overall) and DeepSeek V4 Pro have closed the frontier gap—Kimi K3 ranks ahead of all proprietary models except Claude Fable 5 and GPT-5.6 Sol, and MiniMax M3 scores 80.5% on SWE-bench Verified matching frontier proprietary performance on coding tasks.