ComputeSDK Catalogs 1,429 AI Models Across 12 Gateways and Compares Pricing

Summary

ComputeSDK publishes a weekly catalog of the models advertised by public model-list endpoints from 12 AI gateways. The snapshot contains 1,429 models, with gateway totals ranging from 442 models on OpenRouter to 34 on Neon. The matrix records model coverage, context windows, and reported prices, normalized to U.S. dollars per 1 million tokens where token pricing applies. It also includes models from vendors such as OpenAI, Anthropic, DeepSeek, Google, Moonshot, MiniMax, and Z.ai. The page reports examples including GPT-5 with a 400K context window and a lowest listed price of $1.25 per million input tokens, and GPT OSS 20B with a lowest listed input price of $0.03. Coverage varies by model: several leading entries appear on 11 or 12 gateways, while others appear on fewer. The catalog is not a latency or throughput benchmark, and a listed model without a price is not treated as free. ComputeSDK makes one authenticated GET request to each gateway's `/v1/models` endpoint or equivalent, recording status, response time, model metadata, and pricing. Model IDs are heuristically canonicalized by removing creator prefixes, normalizing separators, and stripping trailing dates, while suffixes such as batch or thinking are retained. The site warns that gateway prices can change before the next weekly index, so users should verify current rates directly before making decisions.