Anthropic's flagship Claude Opus model, carrying the highest relative cost in expensive sessions.
Arena.ai's Agent Arena analysis finds that Gemini 4 Carbon becomes increasingly cost-competitive as task complexity rises. On typical tasks, GPT-6 Astra costs about 3.4x and Claude Opus 5.5 about 3.9x as much as Carbon, while across the most expensive 5% of sessions Opus rises to 4.2x and Astra's cost roughly doubles relative to Carbon. The finding is relevant for budgeting long-horizon agent workloads.
Anthropic launched Claude Fable 5.2 and Claude Mythos 5.2, built around one model with two safeguard levels: Fable 5.2 runs classifiers that block high-risk cyber and research-biology requests and routes them to Opus models, while Mythos 5.2 drops those blocks and is limited to vetted organizations. On FrontierCode 2.0, Fable 5.2 scores 49.1% at Xhigh effort versus 45.5% for Opus 5.5 and 42.8% for GPT-6 Astra, and at High effort it matches Opus 5.5's best score at roughly half the cost per task, though Gemini 4 Carbon still leads AutomationBench and all figures are self-reported. Pricing is $8/$40 per million tokens with a 1M-token context window and 128K output, plus a fast mode up to 2.5x faster.
Anthropic released Claude Fable 5.2 (xHigh), which took first place on Arena's Code Arena WebDev leaderboard with 1832 points, ahead of Claude Opus 5.5, Gemini 4 Carbon and GPT-6 Astra. The update is up to 30% faster than Fable 5.1 and cheaper at $8/$40 per million tokens. Arena also showcased 17 generations spanning 3D worlds, playable games, website UI, data visualizations and interactive code-generated animations.
A new Apple paper introduces AMP1, a harness that lets general-purpose frontier models such as Claude Opus 5.5 and Gemini 4 Argon control a robot arm in a closed loop using camera views, code execution and direct motion commands, with no demonstrations or task-specific training. It achieves 85.6% zero-shot success on RoboLab-120 versus 49.2% for the best VLA and 42.9% for the best world-action model, and succeeds in 104 of 108 trials on a real Franka arm, though the robot idles much of each episode waiting on model inference.
Anthropic released Claude Fable 5.2, which takes the top spot on the Artificial Analysis Intelligence Index v5 at 65, five points ahead of Gemini 4 Carbon and six ahead of Claude Opus 5.5, driven by gains in scientific reasoning, knowledge and FrontierCode 2.0 coding. Pricing drops 20% versus Fable 5.1 to $8/$40 per 1M input/output tokens, though max-effort cost per task of about $26.5 remains far above Opus 5.5.
Anthropic launched Claude Fable 5.2 and Claude Mythos 5.2, its top-tier models, claiming roughly a one-tier gain over Opus 5.5 across nearly all tested capabilities, up to 30% faster output, and API pricing cut from $10/$50 to $8/$40 per million input/output tokens. Fable and Mythos are the same model differing only in guardrails: Fable is open to all paid users, while Mythos is restricted to vetted cybersecurity and life-science institutions. Pro subscribers get temporary plan-quota access to Fable from November 9 to 16.
Prominent AI observer Ethan Mollick tested Anthropic's new Claude Fable 5.2 ahead of public launch and found it highly impressive, building on the already strong Opus 5.5. He frames AGI as matching most human experts on most computer-based intellectual tasks and asks whether this model and imminent OpenAI and Google rivals now cross that limited threshold.
OpenAI has quietly rolled out GPT-6.2 Sol in a gray-release, with users reporting a sharp capability jump after being auto-routed to the new model. It can generate a full Three.js 3D interactive scene from a single prompt and bundle it into one HTML file, while debuting atop the latest Artificial Analysis leaderboard as a direct rival to Claude Opus 5.5.
Developer Simon Willison used Claude Opus 5.5 in a standard chat session to port a Three.js 3D scene into dependency-free vanilla JavaScript and WebGL, with both versions rendering successfully. It is an anecdotal hands-on demo showing the model can rewrite 3D graphics code across libraries, but offers no quantitative benchmark or reproducible methodology.