Developer distribution
Ranking bands are mutually exclusive, avoiding cumulative double-counting.
Reading the results
GPT-6 Astra leads on public evidence for general intelligence, end-to-end computer use, long context, and tool support. Its price keeps the leading margin from becoming larger.
Claude Fable 5.1 is the strongest generally available Claude model. It and Claude Opus 5 remain separate entries, while Claude Mythos 5.1 is not separately ranked because it remains restricted to trusted-access programs.
Gemini 3.8 Flash replaces 3.7 Flash as Google's newest Flash model. Its 1M context, native multimodal I/O, computer-use preview, and public API availability materially improve its rank.
Qwen3.8-Max, GLM-5.3, and Kimi K3 remain the strongest currently evidenced Chinese frontier entries. DeepSeek V4.1 Flash replaces the previous V4 Flash and V4 Pro entries following DeepSeek's public-service migration.
GPT-5.6 Sol remains in the scorecard as a separately accessible, materially different model despite GPT-6 Astra becoming OpenAI's current flagship.
Developer five-dimension heatmap
Darker cells indicate a higher top score achieved by a model from that developer.
Dimension relationship
Switch either axis to explore relationships between strengths; color represents developer.
Category leaders · Top 10
Switch dimensions to see the leading models within the current data scope.