Model Comparison

Pricing updated daily · Last refreshed Aug 31, 2026

Context windows, output limits and per-million-token pricing across providers. Sort any column; filter to the providers you actually ship on.

Filters:
ModelProviderContext windowInput $/1MOutput $/1MMax outputMultimodal
Llama 3.2 1B InstructMeta60,000$0.03$0.2054,000
Command R7B (12-2024)Cohere128,000$0.04$0.154,000
Llama 4 ScoutMeta1,310,720$0.11$0.348,192
Command R (08-2024)Cohere128,000$0.15$0.604,000
Ministral 3 8B 2512Mistral262,144$0.15$0.15209,715
Mistral Small 4Mistral262,144$0.15$0.60209,715
Llama Guard 4 12BMeta163,840$0.18$0.1816,384
Llama 4 MaverickMeta1,048,576$0.20$0.70115,200
Ministral 3 14B 2512Mistral262,144$0.20$0.20209,715
GPT-5.6 LunaOpenAI1,050,000$0.20$1.20128,000
GPT-5.6 Luna ProOpenAI1,050,000$0.20$1.20128,000
Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image)Google65,536$0.25$1.5058,982
Gemini 3.5 Flash LiteGoogle1,048,576$0.30$2.5065,536
Devstral 2 2512Mistral262,144$0.40$2.00209,715
Nano Banana 2 (Gemini 3.1 Flash Image)Google131,072$0.50$3.0032,768
Llama 3.3 70B InstructMeta131,072$0.71$0.71115,200
Gemini 3.6 FlashGoogle1,048,576$0.75$3.7565,536
Gemini 3.7 FlashGoogle1,048,576$0.75$3.7565,536
Mistral Medium 3.5Mistral262,144$1.50$7.50209,715
Claude Sonnet 5Anthropic1,000,000$2.00$10.00128,000
GPT-5.6 Sol ProOpenAI1,050,000$2.00$10.00128,000
GPT-5.6 TerraOpenAI1,050,000$2.00$12.00128,000
GPT-5.6 Terra ProOpenAI1,050,000$2.00$12.00128,000
Command ACohere256,000$2.50$10.008,192
Command R+ (08-2024)Cohere128,000$2.50$10.004,000
Claude Opus 5Anthropic1,000,000$5.00$25.00128,000
Claude Fable 5Anthropic1,000,000$10.00$50.00128,000
Claude Opus 4.8 (Fast)Anthropic1,000,000$10.00$50.00128,000
Claude Opus 5 (Fast)Anthropic1,000,000$10.00$50.00128,000

Showing 29 of 29 models.

Where these numbers come from

Pricing is read from OpenRouter’s public model catalogue once a day. OpenRouter is a routing marketplace, so for a few models its rate can differ from a provider’s own list price — check the vendor’s pricing page before signing a contract on these figures.

Context sensitivity

A large context window is not the same as usable recall. Retrieval accuracy typically degrades well before the advertised limit, so treat the context column as a ceiling rather than a promise.