Self-host it, or buy it per token?
The same work priced three ways on 24 vendors (Alibaba Cloud, Anthropic, AWS, Azure, Cloudflare, Core42, Crusoe, Databricks, DeepInfra, Fireworks AI, Gemini API, Google Cloud, Groq, HF Endpoints, Huawei Cloud, IBM watsonx, Mistral, Nebius, OCI, OpenAI, Snowflake, Telnyx, Together AI, xAI): their APIs per token, rented GPUs, and reserved capacity, and where in the world each one can run it. List prices only, refreshed daily, no vendor's position.
- Self-host or buy: the volume where renting GPUs beats paying per token, for your workload, on each cloud.
- Where it can run: every region a vendor sells a model in, held to a country when the data must stay home.
- Reserved or on demand: PTUs, Bedrock's reserved tier and OCI clusters against per-token prices, at your busy time.
73,758 list prices tracked across 24 vendors, for 89 models that can be compared from one cloud to the next, refreshed every three hours.
What you can do here
- Calculator: What will my workload cost, and where is the break-even?
- Compare: Who sells this model, where, at what price, and on what hardware?
- Ask: Plain-language questions, answered only from the price data.
- Trending: What changed: price moves, models and instances added or withdrawn.
- Data: How many prices each cloud contributes, how fresh they are, and their sources.
Latest changes
- 20 November 2026: GLM 5.3 Flash: output price cut
- 20 November 2026: GLM 5.3 Flash: input price cut
- 20 November 2026: Ultramarine 1 Flash: output price cut
- 20 November 2026: Ultramarine 1 Flash: input price cut
- 18 November 2026: gemini-4-neon launched
- 19 November 2026: Telnyx now sells GPT OSS 120B (Preview) per token from $0.068 in / $0.30 out per 1M