the board members
Providers
Two kinds of sellers set the market: labs that make models, and hosts that serve open-weight models on their own hardware. Prices for the same weights can differ meaningfully between hosts.
Model labs
they train the models they sellOpenAI
28 modelsMaker of the GPT family. The default choice for many production apps, with a wide spread from nano-tier workhorses to frontier reasoning models.
From $0.20/1M input · GPT-5.6 Luna
Anthropic
10 modelsMaker of Claude. Strong on long-context work, coding agents, and careful instruction following, with aggressive prompt-caching discounts.
From $1/1M input · Claude Haiku 4.5
Gemini API models with huge context windows and some of the best price-to-capability ratios in the mid tier.
From $0.25/1M input · Gemini 3.1 Flash-Lite
DeepSeek
3 modelsOpen-weight frontier challenger known for rock-bottom token prices and strong reasoning performance per dollar.
From $0.44/1M input · DeepSeek V4 Flash
xAI
7 modelsMaker of Grok. Fast-shipping frontier lab with real-time X data integration and competitive flagship pricing.
From $1/1M input · Grok Build 0.1
Mistral
7 modelsEuropean lab shipping efficient open and commercial models, popular where data residency and price both matter.
From $0.10/1M input · Ministral 3 3B
Moonshot AI
4 modelsMaker of the Kimi models — open-weight frontier contenders with standout agentic coding results at challenger prices.
From $0.95/1M input · Kimi K2.7 Code
Alibaba Qwen
8 modelsThe Qwen family spans tiny budget models to frontier-class flagships, served via Alibaba Cloud Model Studio.
From $0.05/1M input · Qwen Flash
Z.ai
13 modelsMaker of the GLM series, best known for coding-oriented models and unusually cheap coding subscription plans.
From $0.07/1M input · GLM-4.7-FlashX
MiniMax
6 modelsShanghai-based lab with long-context open-weight models priced for high-volume workloads.
From $0.30/1M input · MiniMax M3
Cohere
7 modelsEnterprise-focused lab behind the Command family, with strengths in RAG, tool use, and multilingual work.
From $0.04/1M input · Command R7B
Perplexity
4 modelsSonar API models that pair generation with built-in web search — priced per token plus search fees.
From $1/1M input · Sonar
Inference hosts
they serve open-weight modelsGroq
7 modelsCustom LPU hardware serving open models at very high speed. Often the cheapest way to run Llama-class models fast.
From $0.03/1M input · Llama Prompt Guard 2 22M
Together AI
16 modelsBroad catalog of open-weight models on serverless endpoints — a one-stop shop for Llama, DeepSeek, and Qwen inference.
From $0.05/1M input · GPT-OSS 20B