Fable 5 is doing something wild on our FrogsGame post-training task. It trains a weaker model to solve the puzzle, peaks at 68%, and produces the only ~10x improvement we see across the benchmark. It spent 17 hours, 25M tokens without human in sight. 34% pass@1, while every other frontier model averages under 4%. We will publish a more detailed analysis soon.

GLM 5.2 is 5x cheaper than Opus 4.8 and 11x than Fable 5, yet it tops PostTrainBench. That’s exciting because lower costs make personalized intelligence economically viable. Every company and country should be able to own models trained on its own data and have sovereignty over it. The future is millions of models, each crafted around the data, values, and decisions of the people who rely on them.

Thank you to our friends at @OpenAI for featuring PostTrainBench in the new model card!

New #1 on PostTrainBench: GLM 5.2 (Max reasoning) hits 34.29%, narrowly beating Opus 4.8 Max (34.08%) What makes GLM 5.2 interesting: zero failed runs across 84 runs (vs ~10% failure rate for Opus agents). The most reliable agent we've seen Leaderboard: posttrainbench.com

A musician gets better by playing, listening, and adjusting. AI is beginning to improve in a similar way: learning from feedback, evaluating its own performance, and refining how it behaves. We call the deliberate shaping of that process modelcrafting. Modelcrafting is the craft of deciding what a model becomes: its character, its judgment, what it pays attention to, and how it improves. It means giving people the ability to shape AI intentionally around their values, their expertise, and the realities of the environments in which it operates. We are building toward responsible recursive self-improvement in the real world: not AI improving itself in isolation, but AI becoming better through feedback from the people it serves. A team should be able to create a model attuned to its judgment, its customers’ needs, its domain’s sensitivities, and the human stakes behind its most consequential decisions. The future should not be defined by models improving on their own, but by people gaining the power to shape the intelligence they depend on. Machine systems should be able to improve without ever losing the human hand on the instrument.

On Claude Fable 5's posttraining capabilities:

models are becoming better at posttraining. Fable 5 specifically solves many of the shortcomings we discussed in our previous report. more mature decision making, questioning the default approaches and better calibration. it feels like a model that "gets it" as @karpathy mentioned. There is still room for improvement specifically understanding the big picture and better calibration but frogsgame is not a good benchmark for that. we need better real world post-training tasks. read more about what we found here.

That said, we dislike FrogsGame as a task internally. The frogs know what they did. We're now sprinting toward adding more useful, real-world posttraining tasks, partly out of ambition, partly to put a distance between us and the frogs 🐸

Opus 4.8 outperforms every other model on AttuneBench - best at picking the response humans actually preferred - biggest MSCEIT four-branch jump of any Opus generation - entire pairwise top-4 is now Anthropic models. non-Anthropic frontiers stall ~50%

We tested 11 frontier LLMs on 200 real human–AI conversations to measure emotional intelligence The result that surprised us: EQ doesn't scale with size or recency. Claude Haiku 4.5 beats Sonnet 4.6. Opus 4.6 performs better than 4.7 It's an orthogonal capability and labs aren't optimizing for it

In post-training, we've learned that once a behavior is measurable, you can train AI to excel at it. EQ is one of the hardest things to verify. AttuneBench makes it measurable through observable signals: whether a model notices distress, tracks shifting preferences, adapts to context, and responds in a way people experience as helpful.

1/ Today we're releasing AttuneBench, the first open EQ benchmark grounded in real multi-turn human-model conversations, scored against what the person actually felt and wanted at each turn. Built by the research team at @pareto_ai in collaboration with @thoughtfullab. Most existing EQ benchmarks rely on: - synthetic prompts - single-turn interactions - third-party annotation None directly measure how a model reads and responds to a real person across a full conversation. We evaluated 11 leading models from major providers, across 200 conversations and 50,000+ first-person annotations.