What each inference provider charges
29 prices from 14 providers, in dollars per million tokens, each read off the provider's own page on the date in the last column. One row per provider and model. Click a price header to sort.
| Provider | Model | Context | Region | Checked | ||
|---|---|---|---|---|---|---|
| Anthropic | fable-5.2 | $8.00 | $40.00 | 1,000,000 | us | |
| Anthropic | opus-5.5 | $4.00 | $20.00 | 1,000,000 | us | |
| Anthropic | sonnet-5.5 | $2.00 | $10.00 | 1,000,000 | us | |
| Anthropic | haiku-5.5 | $0.10 | $0.50 | 1,000,000 | us | |
| Cerebras | gpt-oss-120b | $0.35 | $0.75 | 131,072 | us | |
| Cerebras | qwen-3.8-27b | $0.99 | $1.49 | 131,072 | us | |
| DeepInfra | deepseek-v4.1-flash-1017 | $0.20 | $0.60 | 1,048,576 | us | |
| DeepSeek | deepseek-v4.1-flash-1017 | $0.30 | $1.20 | 1,000,000 | cn | |
| DeepSeek | deepseek-v4-pro-0813 | $1.32 | $3.96 | 1,000,000 | cn | |
| Fireworks | deepseek-v4.1-flash-1017 | $0.30 | $1.20 | 1,048,576 | us | |
| Fireworks | kimi-k3.1-pro | $2.00 | $10.00 | 1,048,576 | us | |
| gemini-4-argon | $2.00 | $10.00 | 1,048,576 | us | ||
| gemini-3.8-flash | $0.75 | $3.75 | 1,048,576 | us | ||
| gemini-3.5-flash-lite | $0.30 | $2.50 | 1,048,576 | us | ||
| Groq | gpt-oss-120b | $0.15 | $0.60 | 131,072 | us | |
| Groq | qwen3.8-27b | $0.80 | $4.00 | 131,072 | us | |
| Mistral | mistral-large-4 | $0.70 | $2.10 | 1,000,000 | eu | |
| Mistral | mistral-small-4 | $0.15 | $0.60 | 256,000 | eu | |
| OpenAI | gpt-6-astra | $10.00 | $50.00 | 1,050,000 | us | |
| OpenAI | gpt-6.1-sol | $2.00 | $10.00 | 1,050,000 | us | |
| OpenAI | gpt-6.1-terra | $1.00 | $5.00 | 1,050,000 | us | |
| OpenAI | gpt-6-luna | $0.10 | $0.50 | 1,050,000 | us | |
| OpenRouter | gpt-6-astra | $10.00 | $50.00 | 1,050,000 | us | |
| Primalabs | deepseek-v4.1-flash-1017 | $0.24 | $0.96 | 1,048,576 | us | |
| Primalabs | mimo-v2.6-pro | $3.50 | $7.00 | 1,048,576 | us | |
| Together | deepseek-v4.1-flash-1017 | $0.30 | $1.20 | 1,000,000 | us | |
| Together | kimi-k3 | $3.00 | $15.00 | 1,050,000 | us | |
| Venice | deepseek-v4-flash-0731 | $0.13 | $0.26 | 1,000,000 | us | |
| xAI | grok-4.7 | $2.00 | $6.00 | 500,000 | us |
Region is where the provider runs inference. A rate marked introductory or off-peak says so under the provider below.
What each provider is for
Anthropic
Closed models in four tiers, all with a million-token window. Pick it for long documents and code review, where the top tier earns its rate on the hardest call and Haiku carries the volume.
fable-5.2: standard tieropus-5.5: standard tiersonnet-5.5: standard tierhaiku-5.5: first 100k input tokens, $0.50 and $2.50 above that
Source: www.anthropic.com/pricing, checked 10 November 2026.
Cerebras
Wafer-scale silicon serving a short list of open models. The reason to be here is time to first token, not the catalogue.
gpt-oss-120b: developer tier, free tier is 65k tokensqwen-3.8-27b: developer tier, free tier is 64k tokens
Source: www.cerebras.ai/pricing, checked 10 November 2026.
DeepInfra
Open-weight models at low list prices, with spot and batch discounts on top. A cheap default for work that can wait in a queue.
deepseek-v4.1-flash-1017: list price, a 51 percent discount runs today at $0.098 and $0.294
Source: deepinfra.com/pricing, checked 10 November 2026.
DeepSeek
The lab's own API for DeepSeek weights, hosted in China and priced in peak and off-peak bands. Direct access, at the source.
deepseek-v4.1-flash-1017: peak rate, off-peak is halfdeepseek-v4-pro-0813: peak rate, off-peak is $0.66 and $1.98
Source: api-docs.deepseek.com/quick_start/pricing, checked 10 November 2026.
Fireworks
Open-weight serving in three service levels: standard, spot and batch. One vendor for several open families, with a price for each level of patience.
deepseek-v4.1-flash-1017: standard serverless, spot is 30 percent off and batch 50 percentkimi-k3.1-pro: standard serverless, priority is $3.00 and $15.00
Source: fireworks.ai/pricing, docs.fireworks.ai/serverless/pricing, checked 10 November 2026.
Gemini models with introductory rates that step up on a published date. Long context and multimodal input are the draw.
gemini-4-argon: introductory rate to 30 November 2026, then $4.00 and $20.00gemini-3.8-flash: introductory rate to 31 December 2026, then $1.50 and $7.50gemini-3.5-flash-lite: paid tier, no introductory discount
Source: ai.google.dev/gemini-api/docs/pricing, checked 10 November 2026.
Groq
Custom silicon for a short list of open models, billed per token. Use it when the first token has to arrive fast.
gpt-oss-120b: developer tierqwen3.8-27b: developer tier
Source: console.groq.com/docs/models, checked 10 November 2026.
Mistral
Mistral's own models, hosted in the EU, with rates on the model card. Data residency is the reason to pick it.
mistral-large-4: price stated on the model cardmistral-small-4: price stated on the model card
Source: docs.mistral.ai/models/model-cards/mistral-large-4, docs.mistral.ai/models/model-cards/mistral-small-4-0-26-03, checked 10 November 2026.
OpenAI
Four tiers from Luna to Astra, with the top tier priced like the frontier it is. The default for tool calling and anything that has to work on the first try.
gpt-6-astra: standard tiergpt-6.1-sol: standard tier, batch is halfgpt-6.1-terra: standard tiergpt-6-luna: standard tier
Source: platform.openai.com/docs/pricing, checked 10 November 2026.
OpenRouter
A router in front of every other provider, at the model's list price plus a fee on credits. Use it to compare providers without four accounts, and to fail over between them.
gpt-6-astra: routing is billed at list price, which runs $10.00 to $60.00 input across the providers that serve this model; credits carry a 5.5 percent fee on Standard and 8 percent on Business
Source: openrouter.ai/pricing, checked 10 November 2026.
Primalabs
A small catalogue of open models, billed per token with reasoning counted as output. Built for agents whose bill has to be predictable.
deepseek-v4.1-flash-1017: cached input is $0.0048 per millionmimo-v2.6-pro: reasoning tokens bill as output
Source: docs.primalabs.ai/inference/models, checked 10 November 2026.
Together
Open-weight hosting with a wide catalogue and one price per model. A second source for an open model you already run elsewhere.
deepseek-v4.1-flash-1017: serverlesskimi-k3: serverless
Source: www.together.ai/pricing, checked 10 November 2026.
Venice
Private inference: prompts are not stored, and the catalogue mixes open models with a few frontier ones. Pick it when retention, not price, is the constraint.
deepseek-v4-flash-0731: the newest DeepSeek Flash Venice lists is the 0731 build; private mode, requests are not retained
Source: docs.venice.ai/overview/pricing, checked 10 November 2026.
xAI
Grok models behind one API, with search over X. Use it when the answer depends on what was posted in the last hour.
grok-4.7: standard tier
Source: docs.x.ai/docs/models, checked 10 November 2026.
Last checked 10 November 2026. Every row links to the page its price came from.