Lambda Inference API (hosted chat completions)

USD per 1M TOKENS - NULL because the service is being wound down and no public token-rate card remains reachable. On 2026-10-15 lambda.ai/inference carries only the notice 'As the Inference API winds down, you can continue deploying and scaling models seamlessly on NVIDIA GPU instances', and docs.lambda.ai/public-cloud/inference-api/ 404s. Lambda's token API was a real OpenAI-compatible product (api.lambda.ai/v1, Llama/DeepSeek/Hermes/Qwen open-weight models) but the per-1M prices that were published are no longer on a reachable page, so quoting a number here would be second-hand. The record exists to answer 'does Lambda rent tokens' (yes, but shutting down) - the GPU-hour side is unaffected and lives in data/supply/.

Recorded deprecated on purpose: a 'where do I run X' tool is more useful when it says 'Lambda's token API is winding down, use their GPU instances or another hoster' than when it silently omits Lambda. Lambda is steering users to on-demand NVIDIA GPU instances (supply/lambda-gpu-cloud-h100, H100 SXM $3.99/GPU-hr) - the rent-the-hardware path the supply type already covers. models_served reflects the open-weight catalog the API advertised (llama-3.3-70b, deepseek-r1, hermes-3, qwen3-32b class); entries are repo model slugs, indicative not re-verified against the live endpoint which is being retired. If the wind-down completes this record can be dropped; until then it is the honest answer to the coverage question.