LLM API Price Comparison: OpenAI, Claude, Gemini, Grok, DeepSeek, Qwen, Kimi, GLM and More (October 2026)
TL;DR: Official per-million-token list prices for the major API models in two sourced tables; the cheapest models cost a small fraction of flagships, and caching, long-prompt tiers and batch discounts change the real bill.
Two tables, 32 models, 13 vendors — every number copied by hand from the vendor’s official pricing page, with a link to the source. The first table (OpenAI, Anthropic, Google, DeepSeek, Qwen) was checked on 2026-10-18; the second (Grok, Mistral, Kimi, GLM, Doubao Seed, Hunyuan, MiniMax, MiMo) was added on 2026-10-21. Below them: the fine print that changes what you actually pay.
Compare cost for your own prompt →All guides
Standard prices, checked 2026-10-18
USD per 1 million tokens, standard (non-batch, non-cached) tier. “In” is input, “out” is output.
| Vendor | Model | In | Out | Notes |
|---|---|---|---|---|
| OpenAI | gpt-6-astra | $10.00 | $50.00 | Short-context rate |
| OpenAI | gpt-6.1-sol | $2.00 | $10.00 | Short-context rate |
| OpenAI | gpt-6-luna | $0.10 | $0.50 | Short-context rate |
| OpenAI | gpt-5.3-codex | $1.75 | $14.00 | |
| Anthropic | Claude Fable 5.1 | $10.00 | $50.00 | |
| Anthropic | Claude Opus 5.5 | $4.00 | $20.00 | |
| Anthropic | Claude Sonnet 5.5 | $2.00 | $10.00 | |
| Anthropic | Claude Haiku 5.5 | $0.10 | $0.50 | Prompts ≤100K tokens; above: $0.50 / $2.50 |
| gemini-4-argon | $2.00 | $10.00 | Through 2026-11-30; $4 / $20 from 2026-12-01 | |
| gemini-3.8-flash | $0.75 | $3.75 | Through 2026-12-31; $1.50 / $7.50 from 2027-01-01 | |
| gemini-3.5-flash-lite | $0.30 | $2.50 | ||
| DeepSeek | deepseek-v4-pro | $1.32 | $3.96 | Peak, cache miss; off-peak $0.66 / $1.98 |
| DeepSeek | deepseek-flash | $0.30 | $1.20 | Peak, cache miss; off-peak $0.15 / $0.60 |
| Alibaba | qwen3.8-max | $2.00 | $6.00 | Model Studio International (Singapore) |
| Alibaba | qwen3.7-plus | $0.40 | $1.60 | List price ≤256K; above: $1.20 / $4.80; page shows a limited-time 20% off |
| Alibaba | qwen3.8-flash | $0.15 | $0.47 | Model Studio International (Singapore) |
Sources: OpenAI pricing · Anthropic pricing · Gemini API pricing · DeepSeek pricing · Alibaba Cloud Model Studio pricing. Prices change often; always confirm on the vendor page before budgeting.
More vendors, checked 2026-10-21
The flagship and the main low-cost model from eight more vendors, in the same units: USD per 1 million tokens, standard real-time tier, international list price. Chinese vendors also sell in CNY on their mainland consoles at different prices.
| Vendor | Model | In | Out | Notes |
|---|---|---|---|---|
| xAI | grok-4.7 | $2.00 | $6.00 | Prompts <200K; ≥200K: $4 / $12 |
| xAI | grok-4.3 | $1.25 | $2.50 | Prompts <200K; ≥200K: $2.50 / $5 |
| Mistral | mistral-medium-3-5 | $1.50 | $7.50 | Model-card price; 256K context |
| Mistral | mistral-small-2603 | $0.15 | $0.60 | Model-card price (Mistral Small 4) |
| Moonshot | kimi-k3 | $3.00 | $15.00 | Cache hit $0.30; cache writes billed separately |
| Moonshot | kimi-k2.6 | $0.95 | $4.00 | Cache hit $0.16 |
| Zhipu (Z.ai) | glm-5.3 | $1.40 | $4.40 | Z.ai international; cache hit $0.26 |
| Zhipu (Z.ai) | glm-5.3-flash | $0.15 | $0.50 | Z.ai international; cache hit $0.03 |
| ByteDance | seed-2-0-pro | $0.50 | $3.00 | BytePlus ModelArk, prompts ≤128K; above: $1 / $6 |
| ByteDance | seed-2-0-mini | $0.10 | $0.40 | BytePlus ModelArk, prompts ≤128K; above: $0.20 / $0.80 |
| Tencent | hy4-preview | $0.834 | $2.501 | TokenHub international |
| Tencent | hy3 | $0.132 | $0.528 | TokenHub international |
| MiniMax | MiniMax-M3 | $0.30 | $1.20 | Standard tier, prompts ≤512K; above: $0.60 / $2.40 |
| MiniMax | MiniMax-M3.1-Flash | $0.15 | $0.60 | Cache read $0.015 |
| Xiaomi | mimo-v2.6-pro | $0.435 | $0.87 | Overseas USD price; China ¥3 / ¥6 |
| Xiaomi | mimo-v2.6-flash | $0.14 | $0.28 | Overseas USD price; China ¥1 / ¥2 |
Sources: xAI pricing · Mistral Medium 3.5 · Mistral Small 4 · Kimi API pricing · Z.ai GLM pricing · BytePlus ModelArk pricing · Tencent TokenHub prices · MiniMax pricing · Xiaomi MiMo pricing. Doubao’s Seed models are listed under their BytePlus IDs (seed-2-0-pro-260328, seed-2-0-mini-260428); Mistral’s pages do not publish a fixed cached-input price.
Same workload as below (20M input + 5M output tokens a month), priced on these sixteen rows:
| Model | Monthly cost |
|---|---|
| kimi-k3 | $135.00 |
| grok-4.7 | $70.00 |
| mistral-medium-3-5 | $67.50 |
| glm-5.3 | $50.00 |
| kimi-k2.6 | $39.00 |
| grok-4.3 | $37.50 |
| hy4-preview | $29.19 |
| seed-2-0-pro | $25.00 |
| mimo-v2.6-pro | $13.05 |
| MiniMax-M3 | $12.00 |
| mistral-small-2603 | $6.00 |
| MiniMax-M3.1-Flash | $6.00 |
| glm-5.3-flash | $5.50 |
| hy3 | $5.28 |
| mimo-v2.6-flash | $4.20 |
| seed-2-0-mini | $4.00 |
How to read this table
Output dominates. In every row, output costs roughly 3× to 8× input. A chat assistant that writes long answers is mostly paying for output; a classifier that reads a long document and returns one label is mostly paying for input. Pick the model by the column your workload actually stresses.
Per-token prices are not per-text prices. Each vendor uses its own tokenizer. Anthropic states that Claude 4.7 and later produce about 30% more tokens for the same text than earlier Claude models, so a Claude model and a GPT model with identical per-token prices will not cost the same for the same prompt. Our token counter applies a ×1.3 estimate to Claude rows for exactly this reason.
“Cheapest” depends on the shape of the request. With 2,000 input and 500 output tokens per call, Claude Sonnet 5.5 and gpt-6.1-sol both come to $0.009 per request, gemini-3.8-flash to about $0.0034, deepseek-flash (peak) to $0.0012 and qwen3.8-flash to about $0.0005. Double the output and the ranking between mid-range models can shift.
The fine print that changes your bill
Scheduled price changes (Google)
Google’s pricing page lists gemini-4-argon at $2 / $10 through 2026-11-30, doubling to $4 / $20 from 2026-12-01, and gemini-3.8-flash at $0.75 / $3.75 through 2026-12-31, doubling to $1.50 / $7.50 from 2027-01-01. If you are budgeting a quarter ahead, use the later price.
Peak and off-peak (DeepSeek)
DeepSeek’s off-peak rates are half of peak. Its pricing page defines peak hours as 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday, excluding Chinese public holidays — that is 09:00–12:00 and 14:00–18:00 Beijing time. Everything else, including weekends, is off-peak. Batch-style jobs scheduled for evenings or weekends pay half. DeepSeek also lists much lower cache-hit input prices (for example $0.006 per million at peak for deepseek-flash).
Prompt-length tiers
Claude Haiku 5.5 is priced by prompt length: a request whose prompt exceeds 100,000 tokens pays $0.50 / $2.50 instead of $0.10 / $0.50, and Anthropic counts cache reads and writes toward that length. qwen3.7-plus moves to $1.20 / $4.80 above 256K input tokens. OpenAI’s GPT-6 rows above are the short-context rates.
Region (Qwen)
Alibaba Cloud Model Studio publishes different prices per deployment region. The rows above are the International (Singapore) prices in USD. If you use the mainland China console, check the CNY price list for your region.
Discounts that stack
- Batch: OpenAI’s Batch API and Anthropic’s Message Batches are both documented at 50% off input and output, in exchange for asynchronous processing (OpenAI: completion within 24 hours).
- Prompt caching: repeated prefixes are billed at a fraction of the input price on all five vendors — see our caching guide for the multipliers.
One workload, every model
To make the table concrete, here is the same workload priced on every row: 10,000 requests a month, each with 2,000 input and 500 output tokens (20M input + 5M output tokens), at the standard rates above. Token counts are taken as equal across models, which flatters Claude 4.7+ models by roughly 30%.
| Model | Monthly cost |
|---|---|
| gpt-6-astra / Claude Fable 5.1 | $450.00 |
| Claude Opus 5.5 | $180.00 |
| gemini-4-argon (from 2026-12-01) | $180.00 |
| gpt-5.3-codex | $105.00 |
| gpt-6.1-sol / Claude Sonnet 5.5 | $90.00 |
| gemini-4-argon (through 2026-11-30) | $90.00 |
| qwen3.8-max | $70.00 |
| deepseek-v4-pro (peak) | $46.20 |
| gemini-3.8-flash (through 2026-12-31) | $33.75 |
| deepseek-v4-pro (off-peak) | $23.10 |
| gemini-3.5-flash-lite | $18.50 |
| qwen3.7-plus | $16.00 |
| deepseek-flash (peak) | $12.00 |
| deepseek-flash (off-peak) | $6.00 |
| qwen3.8-flash | $5.35 |
| gpt-6-luna / Claude Haiku 5.5 | $4.50 |
The spread is about 100× from the most to the least expensive row. That is why routing — small models for easy requests, large models only where they earn their price — usually saves more than haggling over a few cents per million.
What these prices do not include
- Tool fees. Built-in web search and similar tools are often billed per call. DeepSeek, for instance, prices its Search and Fetch APIs at $0.75 per 1,000 successful requests, and says server-side tool calls made through DeepSeek Harness, Claude Code or Codex start being billed at 00:00 Beijing time on 1 November 2026.
- Images, audio and files. Multimodal input is converted to tokens with its own rules; count those with the vendor’s token-counting endpoint.
- Taxes and currency. All prices here are USD before tax; mainland China consoles bill in CNY with their own price lists.
A sensible way to choose
- Collect 20–50 real requests from your app, with their outputs.
- Shortlist two or three models per price band that pass your quality bar on those requests.
- Paste a typical request into the token counter, set expected output tokens and monthly requests, and compare the totals.
- Add caching and batch savings where they apply, then re-check the vendor page on the day you decide.
Sources
Vendor features and prices were checked against these official pages on 2026-10-21. Products change; if something here no longer matches, the official page is right.
- OpenAI — API pricing
https://developers.openai.com/api/docs/pricing - Anthropic — Pricing
https://docs.anthropic.com/en/docs/about-claude/pricing - Google — Gemini API pricing
https://ai.google.dev/gemini-api/docs/pricing - DeepSeek — Models & Pricing
https://api-docs.deepseek.com/quick_start/pricing - Alibaba Cloud — Model Studio model pricing
https://www.alibabacloud.com/help/en/model-studio/model-pricing - OpenAI — Batch API
https://developers.openai.com/api/docs/guides/batch - xAI — API pricing
https://docs.x.ai/developers/pricing - Mistral — Mistral Medium 3.5 model card
https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04 - Mistral — Mistral Small 4 model card
https://docs.mistral.ai/models/model-cards/mistral-small-4-0-26-03 - Moonshot — Kimi API model pricing
https://platform.kimi.ai/docs/pricing/chat - Z.ai — GLM pricing
https://docs.z.ai/guides/overview/pricing - BytePlus — ModelArk pricing (Seed / Doubao)
https://docs.byteplus.com/en/docs/ModelArk/1544106 - Tencent Cloud — TokenHub model prices (Hunyuan)
https://intl.cloud.tencent.com/zh/document/product/1300/78937 - MiniMax — Pricing overview
https://platform.minimax.io/docs/guides/pricing-paygo - Xiaomi — MiMo pay-as-you-go pricing
https://mimo.mi.com/docs/en-US/price/pay-as-you-go