Monthly bill per host · lower is better
30 days, at each host’s sourced price
Calculators on the same sourced prices as the index
Tools
Rankings compare. These compute: the bill for your own volumes, on the same sourced prices as the rest of the index. The cost calculator prices your workload at 864 endpoints.
LLM cost calculator
864
priced endpoints
Cost of a whole stack
15
non-LLM providers priced
Tokenizer tax
21
tokenizer families measured
Self-host an LLM
95
open models costed
LLM cost calculator
What your workload costs
Your tokens per request and your volume, priced at every host of the model. 112 models served by two hosts or more; the full calculator covers every endpoint.
Four calculators
Beyond the token bill
Same prices, other questions: every model, the rest of your stack, the tokenizer, your own GPU.
Every model
LLM cost calculator
Your tokens per request and your volume, priced at every host of every model.
864priced endpoints
Open →The rest of the bill
Cost of a whole stack
The LLM bill next to hosting, database, email and payments, for a product you are about to build.
15non-LLM providers priced
Open →Per character, not per token
Tokenizer tax
The same text costs a different number of tokens at each maker: price per million characters, not per million tokens.
21tokenizer families measured
Open →Your own GPU
Self-host an LLM
The GPU a model needs, what it costs per hour, and the volume at which it beats the API.
95open models costed
Open →Run a model locally
One page per open model
The VRAM it needs, the GPU that fits it, and what it costs against the API.
- command-aCohere
- command-rCohere
- deepseek-chatDeepSeek
- deepseek-chat-v3DeepSeek
- deepseek-chat-v3.1DeepSeek
- deepseek-r1DeepSeek
- deepseek-v3DeepSeek
- deepseek-v3.1-terminusDeepSeek
- deepseek-v3.2DeepSeek
- deepseek-v3.2-expDeepSeek
- deepseek-v4-flashDeepSeek
- deepseek-v4-flash-0423DeepSeek
- deepseek-v4-flash-vision-expDeepSeek
- deepseek-v4.1-flashDeepSeek
- deepseek-v4.1-flash-1017DeepSeek
- gemma-3-27bGoogle
- gemma-4-26b-a4bGoogle
- gemma-4-31bGoogle
- glm-4.5-airGLM (Z.ai)
- glm-4.5vGLM (Z.ai)
- glm-4.6GLM (Z.ai)
- glm-4.6vGLM (Z.ai)
- glm-4.7GLM (Z.ai)
- glm-4.7-flashGLM (Z.ai)
- glm-5GLM (Z.ai)
- glm-5.1GLM (Z.ai)
- glm-5.2GLM (Z.ai)
- glm-5.3GLM (Z.ai)
- glm-5.3-flashGLM (Z.ai)
- gpt-oss-120bOpenAI
- gpt-oss-20bOpenAI
- granite-4.2-8bIBM
- hy3Tencent
- inklingThinking Machines
- inkling-smallThinking Machines
- kimi-k2-thinkingKimi (Moonshot)
- kimi-k2.5Kimi (Moonshot)
- kimi-k2.6Kimi (Moonshot)
- kimi-k2.7-codeKimi (Moonshot)
- ling-3.0-flashinclusionAI
- ling-3.0-flash-fininclusionAI
- ling-3.0-flash-vlinclusionAI
- ling-3.1-flashinclusionAI
- llama-3.1-70bMeta
- llama-3.1-8bMeta
- llama-3.2-3bMeta
- llama-3.3-70bMeta
- llama-4-maverickMeta
- llama-4-scoutMeta
- minimax-m1MiniMax
- minimax-m2MiniMax
- minimax-m2.1MiniMax
- minimax-m2.5MiniMax
- minimax-m2.7MiniMax
- minimax-m3MiniMax
- ministral-14bMistral
- ministral-3bMistral
- ministral-8bMistral
- mistral-largeMistral
- mistral-nemoMistral
- mistral-small-3.2-24bMistral
- mistral-small-4Mistral
- mixtral-8x22bMistral
- muse-glimmer-30bMeta
- muse-spark-openMeta
- nemotron-3-nano-30b-a3bNVIDIA
- nemotron-3-super-120b-a12bNVIDIA
- nemotron-3-ultra-550b-a55bNVIDIA
- nemotron-3.5-lightningNVIDIA
- nemotron-3.5-super-vlNVIDIA
- qwen-2.5-72bQwen (Alibaba)
- qwen3-14bQwen (Alibaba)
- qwen3-235b-a22bQwen (Alibaba)
- qwen3-235b-a22b-thinkingQwen (Alibaba)
- qwen3-30b-a3bQwen (Alibaba)
- qwen3-32bQwen (Alibaba)
- qwen3-coderQwen (Alibaba)
- qwen3-coder-30b-a3bQwen (Alibaba)
- qwen3-coder-nextQwen (Alibaba)
- qwen3-next-80b-a3bQwen (Alibaba)
- qwen3-next-80b-a3b-thinkingQwen (Alibaba)
- qwen3-vl-235b-a22bQwen (Alibaba)
- qwen3-vl-235b-a22b-thinkingQwen (Alibaba)
- qwen3-vl-30b-a3bQwen (Alibaba)
- qwen3-vl-30b-a3b-thinkingQwen (Alibaba)
- qwen3-vl-8bQwen (Alibaba)
- qwen3.5-122b-a10bQwen (Alibaba)
- qwen3.5-27bQwen (Alibaba)
- qwen3.5-35b-a3bQwen (Alibaba)
- qwen3.5-397b-a17bQwen (Alibaba)
- qwen3.5-9bQwen (Alibaba)
- qwen3.6-27bQwen (Alibaba)
- qwen3.6-35b-a3bQwen (Alibaba)
- qwen3.8-27bQwen (Alibaba)
- voxtral-small-24bMistral
Open data · CC BY 4.0
Use the data
Build your own tool on the same prices. Free, with a credit line.
The full dataset
Every tracked endpoint with its verified price, source URL and verification date. Regenerated at each rebuild.
GET /api/all.jsonOpen the JSON →Price history
One row per endpoint and date, a point only when the price actually changed.
GET /api/pricing-history.csvDownload the CSV →Benchmark readings, archived
Our weekly speed bench ran until 20 October 2026. Its last readings stay available, frozen, for reference.
GET /api/benchmark-readings.csvDownload the CSV →OpenAPI spec
Machine description of all data endpoints, including per-model JSON and price badges.
GET /openapi.jsonOpen the spec →