Trust & method
Sources & evidence: where ByteCosts pricing comes from
ByteCosts prices come from each provider’s own pricing and documentation pages, normalized into one index and dated. This page lists the primary source for the major providers we track22 of the 193 providers in the indexso any figure can be traced back to where it was read. Provider-vs-third-party precedence is strict: a provider’s pricing page beats its docs page, which beats normalized third-party catalog structure. Each official price row also carries a confidence grade (A+ to E) and a last-checked date. Use these links to verify a current rate before you commit; list prices change, and ByteCosts is refreshed manually rather than live.
Provider source pages
The primary pricing or documentation source for each major provider in the index. These are the pages a price is read from during a manual refresh.
| Provider | Priced models | Source |
|---|---|---|
| Alibaba Qwen | 46 | https://www.alibabacloud.com |
| Amazon Bedrock | 116 | https://aws.amazon.com/bedrock/pricing/ |
| Anthropic | 13 | https://platform.claude.com/docs/en/about-claude/pricing |
| Cloudflare Workers AI | 24 | https://developers.cloudflare.com/workers-ai/platform/pricing/ |
| Cohere | 8 | https://cohere.com/pricing |
| Deep Infra | 59 | https://deepinfra.com/pricing |
| DeepSeek | 3 | https://api-docs.deepseek.com/quick_start/pricing |
| Fireworks AI | 23 | https://fireworks.ai/pricing |
| Google Gemini | 16 | https://ai.google.dev/gemini-api/docs/pricing |
| Google Vertex AI | 37 | https://cloud.google.com/vertex-ai/generative-ai/pricing |
| Groq | 5 | https://groq.com/pricing/ |
| Llama | 0 | https://llama.developer.meta.com |
| MiniMax | 7 | https://platform.minimax.io/docs/guides/pricing-paygo |
| Mistral | 28 | https://docs.mistral.ai/platform/pricing/ |
| Moonshot AI | 10 | https://platform.moonshot.ai |
| OpenAI | 38 | https://developers.openai.com/api/docs/pricing |
| OpenRouter | 319 | https://openrouter.ai/docs#models |
| Perplexity | 4 | https://docs.perplexity.ai/docs/getting-started/pricing |
| Together AI | 36 | https://www.together.ai/pricing |
| Upstage | 4 | https://www.upstage.ai/pricing |
| xAI | 7 | https://docs.x.ai/docs/models |
| Z.AI | 12 | https://docs.z.ai |
How to verify a price
To confirm any rate ByteCosts shows:
- Open the provider’s source page from the table above and find the model.
- Check the model’s confidence grade and last-checked date on its ByteCosts row, A/A+ rows were read from an official page.
- Remember every rate is normalized to USD per 1M tokens; convert if the provider quotes per-1K or per-request.
- Treat list prices as pre-discount: enterprise, volume, and committed-use rates are negotiated and not shown here.
- If a provider has changed a price since the last manual refresh, the source page is authoritative, ByteCosts updates on the next manual pass.
Catalog structure and third-party discovery
The public model catalog is derived in part from models.dev (MIT License: https://github.com/anomalyco/models.dev). List prices on ByteCosts are still attributed to each provider’s first-party pricing or docs URL, models.dev supplies catalog structure, not a substitute for those provider pages.
GPU rental discovery uses the SkyPilot catalog as a discovery aid, not as a licensed catalog dump. Provider APIs and pricing pages remain the ground truth for GPU rates.
ByteCosts does not publish a public Artificial Analysis dataset. Quality-index, Elo, rank, and related third-party benchmark scores are not redistributed on this site or in the public exports.
ByteCosts sources & evidence. ByteCosts. Updated August 22, 2026. https://bytecosts.com/sources/
Sources
- models.dev catalog (MIT)
- models.dev repository
- Alibaba Qwen pricing source
- Amazon Bedrock pricing source
- Anthropic pricing source
- Cloudflare Workers AI pricing source
- Cohere pricing source
- Deep Infra pricing source
- DeepSeek pricing source
- Fireworks AI pricing source
- Google Gemini pricing source
- Google Vertex AI pricing source