输入价格$0.50/百万 tokens
输出价格$1.50/百万 tokens
上下文256Ktokens
P50 延迟0.5s首字延迟厂商提供
吞吐量220tok/s厂商提供
The best open-weight alternative for self-hosted or sovereign deployments. Closed-source frontier still leads on raw quality, but cost-per-token via inference providers like Together / Fireworks is the lowest in the table, and there is no vendor lock-in. Worth the integration work when latency, sovereignty, or fine-tuning flexibility matter more than absolute benchmark performance.
完整定价详情
全部价格按美元 / 百万 token 计 · 所有隐性成本已揭示| Pricing tier | Input | Output | Notes |
|---|---|---|---|
Standard Pay-as-you-go API rate card | $0.50 | $1.50 | Default if no headers set |
With prompt caching Cached read · cache write | $0.05−90% | $1.50 | Cache writes priced at standard input ($0.50); cached reads 90% off. Requires cache_control header. |
Batch API (50% off) 24h SLA, async | $0.25−50% | $0.75−50% | Send up to 50K requests in one batch; results within 24h |
Image input (vision) Per image, low/high detail differ | $0.50 + image fee | $1.50 | $0.50 per 1024×1024 image at high detail; low detail ~$0.10 |
Reasoning tokens Hidden chain-of-thought tokens | $0.50 | $1.50 × ~3-8 | When using reasoning.effort: high, output tokens include hidden reasoning. Real cost can be 3-8× headline output. |
Structured output JSON mode + schema | $0.50 | $1.50 | No extra cost; first call may be slower (~2x latency) due to schema compilation |
能力
Vision (image understanding)Function callingStructured output (JSON schema)Reasoning modeMultilingual (32+ langs)Batch APIAudio I/OFine-tuning