AI developer tools · What shipped

Anthropic introduced Claude Fable 5.1 and Claude Mythos 5.1, positioning them as the world's most advanced models for coding and knowledge work. Fable 5.1 achieves significantly higher performance than its predecessor across multiple benchmarks: on Terminal-Bench 4.0 (agentic coding) it scores 60.9% versus Fable 5's 42.0%, on Humanity's Last Exam it reaches 65.0% with tools versus Fable 5's 63.8%, and on CursorBench 3.2.0 it scores 73.4% versus Fable 5's 70.5%. The model also costs approximately 25% less than Fable 5 for typical workloads due to 75% reduced pricing on cache reads, with savings up to 45% for highly agentic tasks. Fable 5.1 demonstrates particular strength in complex coding tasks—finding root causes of software bugs that other models and human engineers could not identify—and shows improved performance on knowledge work benchmarks including GDPval-AA v2 (1853 versus Fable 5's 1724) and AutomationBench (31.4% versus 26.9%).

Artificial Analysis Intelligence Index v4.3 now incorporates 10 evaluations including AA-Briefcase, AutomationBench-AA, Terminal-Bench v4.0, SciCode, Humanity's Last Exam, and others to measure AI model performance. Gemini 3.8 Flash is positioned in the most attractive quadrant on the Intelligence Index vs. cost analysis, with performance measured across output speed, cost per task, and capability scores spanning finance, legal, healthcare, engineering, economics, and strategy categories.

Meta published official Muse Spark 1.3 benchmarks on 2026-09-02, with the model achieving 75.4% on DeepSWE 1.1 for end-to-end software engineering issue resolution, 88.8% on Terminal-Bench 2.1 for terminal environment navigation, 59.4% on SWEAtlas CodeBase QnA for repository comprehension, and 98.5% on long-context retrieval across its 1M-token context window. The model completes coding workflows using roughly 20% fewer tool calls and 25% fewer tokens than Muse Spark 1.2 while maintaining the full 1M-token context window, with efficiency gains tracked on identical multi-step tasks in Meta's Muse Code CLI agent. Two release variants were launched: Muse Spark 1.3 (max) remains in limited preview for selected Meta partners, while Muse Spark 1.3 (xhigh) is available for general production use through the Meta API and Muse Code. Model weights remain closed for both tiers, with pricing estimated at roughly $0.80 per million tokens, though Meta has not confirmed an itemized rate card.

Alibaba's Open Code Review, an AI-powered code review CLI tool originally developed internally and now open-sourced, published benchmark results comparing its performance against general-purpose agents like Claude Code. Using a real-world benchmark (AACR-Bench) built from 50 open-source repositories, 200 real pull requests across 10 programming languages, and validated by 80+ senior engineers, Open Code Review achieved significantly higher precision and F1 scores while consuming approximately 1/9 of the tokens and completing reviews faster than Claude Code. The tool combines deterministic engineering with agent capabilities, featuring precise file selection, smart file bundling, fine-grained rule matching, and scenario-tuned prompts to address common issues with general-purpose agents like incomplete coverage, position drift, and unstable quality.