The labs

Claude family. Leads measured intelligence and repository coding; premium pricing, strong computer-use and writing.

4 current · 7 tracked · lead Claude Opus 5

GPT-5.6 is a three-tier stack (Sol / Terra / Luna). Sol leads abstract reasoning; Luna is the cheap high-volume option.

4 current · 6 tracked · lead GPT-5.6 Sol

Gemini 3 Flash line is the speed story of 2026. Pro still trails Claude and GPT on agents; multimodal remains a strength.

2 current · 4 tracked · lead Gemini 3.7 Flash

Grok 4.6 ties GPT-5.6 Sol on the Intelligence Index at a third of the output price. 500K context is the main ceiling.

1 current · 2 tracked · lead Grok 4.6

Qwen3.8 Max is the best evidence-backed open-weight flagship. Strong multilingual and reasoning; slower on tool loops.

3 current · 3 tracked · lead Qwen3.8 Max

Kimi K3 is a 2.8T MoE with open weights and near-frontier intelligence. Best near-frontier value on several boards.

1 current · 1 tracked · lead Kimi K3

GLM-5.3 sits one point off the Intelligence Index lead at $1.40 / $4.40. Agentic scores punch above the price.

2 current · 3 tracked · lead GLM-5.3

V4 Pro and Flash are the default self-host / cheap-API pair. Not the intelligence crown, but the cost curve is unmatched.

2 current · 2 tracked · lead DeepSeek V4 Pro

Muse Spark is Meta’s closed frontier line. Llama 4 remains the open-weight workhorse for local and fine-tune stacks.

2 current · 3 tracked · lead Muse Spark 1.2

European lab. Medium 3.5 is competent on coding, not competing for the intelligence crown. Strong EU-data story.

1 current · 1 tracked · lead Mistral Medium 3.5

M3 is an open-weight coding and video-adjacent model. Vendor SWE-bench is strong; independent intelligence is mid-pack.

1 current · 1 tracked · lead MiniMax M3

MiMo-V2.5-Pro is Xiaomi’s first model that shows up on frontier boards. Watch the evidence; coverage is still thin.

1 current · 1 tracked · lead MiMo-V2.5-Pro