By the lab's own price
AI models ranked by price
Flagship and cheap API models, priced from each lab's own page and ranked three ways. Where a score is blank, the openly licensed source had no row we were willing to match by hand.
Blended price: three input tokens for every one output token, per million tokens. Lower is cheaper.
Blended price, in this order
Each bar is the blended price per million tokens, on a log scale so the cheap models still show. Rank 1 is highlighted in amber.
- 1Qwen3.7 Flash$0.055
- 2Ministral 3 3B$0.10
- 3GPT-5 nano$0.1375
- 4Ministral 3 8B$0.15
- 5Claude Haiku 5.5$0.20
- 6GPT-6 Luna$0.20
- 7Qwen3.8 Flash$0.23
- 8GPT-4o mini$0.2625
- 9Mistral Small 4$0.2625
- 10GPT-5.6 Luna$0.45
Score against price
Epoch Capabilities Index against blended price. The dashed line is the value frontier: nothing cheaper scores higher.
Epoch Capabilities IndexBlended price per million tokens, log scale
Value frontierRank 1OpenAIAnthropicMistralAlibaba
Value frontier: Qwen3.7 Flash, GPT-6 Luna, Claude Haiku 5.5
Qwen3.7 Flash · Alibaba · ECI 144.14 · blended $0.055
| Rank | Model | Lab | Input | Output | Blended | ECI | Arena | Note |
|---|---|---|---|---|---|---|---|---|
| 1 | Qwen3.7 Flash | AlibabaListed companyBABA9988.HK | $0.03 | $0.13 | $0.055 | 144.14 | — | International price, Singapore, for prompts up to 32K tokens. From 32K to 256K the page lists $0.10 input and $0.40 output; from 256K to 1M, $0.20 and $0.80. |
| 2 | Ministral 3 3B | MistralListed stakeASML | $0.10 | $0.10 | $0.10 | — | — | |
| 3 | GPT-5 nano | OpenAIListed stakeMSFT | $0.05 | $0.40 | $0.1375 | 139.25 | — | |
| 4 | Ministral 3 8B | MistralListed stakeASML | $0.15 | $0.15 | $0.15 | — | — | |
| 5 | Claude Haiku 5.5 | AnthropicListed stakeGOOGLAMZN | $0.10 | $0.50 | $0.20 | 160.37 | — | This rate is for prompts up to 100,000 tokens. Over that, the page lists $0.50 input and $2.50 output. Claude 4.7 and later use a newer tokenizer, about 30% more tokens for the same text. |
| 6 | GPT-6 Luna | OpenAIListed stakeMSFT | $0.10 | $0.50 | $0.20 | 156.75 | — | Short-context rate, for prompts up to 272K input tokens. Longer prompts are priced higher on the same page. |
| 7 | Qwen3.8 Flash | AlibabaListed companyBABA9988.HK | $0.15 | $0.47 | $0.23 | — | — | International price, Singapore region, on Alibaba Cloud Model Studio. Other regions on the same page differ. |
| 8 | GPT-4o mini | OpenAIListed stakeMSFT | $0.15 | $0.60 | $0.2625 | 126.45 | — | |
| 9 | Mistral Small 4 | MistralListed stakeASML | $0.15 | $0.60 | $0.2625 | — | — | |
| 10 | GPT-5.6 Luna | OpenAIListed stakeMSFT | $0.20 | $1.20 | $0.45 | 156.67 | — | Short-context rate, for prompts up to 272K input tokens. Longer prompts are priced higher on the same page. |
Showing 10 of 32 models · Free with Spectator
| 11 | DeepSeek V4.1 Flash | DeepSeekPrivate, no listed exposure | $0.30 | $1.20 | $0.525 | 158.83 | — | Peak rate. Off-peak is half. Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday, excluding Chinese public holidays. |
| 12 | Gemini 3.1 Flash-Lite | GoogleListed companyGOOGL | $0.25 | $1.50 | $0.5625 | 144.34 | — | Standard paid rate for text. Batch is half. The page states no end date for this price. |
| 13 | Gemini 3.5 Flash-Lite | GoogleListed companyGOOGL | $0.30 | $2.50 | $0.85 | 144.97 | 1435.5 | Standard paid rate for text. Batch is half. The page states no end date for this price. |
| 14 | Mistral Large 4 | MistralListed stakeASML | $0.70 | $2.10 | $1.05 | 150.59 | 1430.1 | Cached input is listed separately and is lower. That discount is not the rate ranked here. |
| 15 | Qwen3.8 27B | AlibabaListed companyBABA9988.HK | $0.50 | $3 | $1.125 | 151.8 | — | International price, Singapore region, on Alibaba Cloud Model Studio. Other regions on the same page differ. |
| 16 | Gemini 3.8 Flash | GoogleListed companyGOOGL | $0.75 | $3.75 | $1.50 | 157.64 | — | Standard paid rate through 31 December 2026, then $1.50 input and $7.50 output. Batch and Flex are half. |
| 17 | Grok 4.20 | xAIWholly owned by the listed companySPCX | $1.25 | $2.50 | $1.5625 | 152.26 | — | Rate below 200,000 prompt tokens. At or above that, the whole request is billed at the long-context rate on the same page. |
| 18 | Grok 4.3 | xAIWholly owned by the listed companySPCX | $1.25 | $2.50 | $1.5625 | — | 1396.9 | Rate below 200,000 prompt tokens. At or above that, the whole request is billed at the long-context rate on the same page. |
| 19 | DeepSeek V4 Pro | DeepSeekPrivate, no listed exposure | $1.32 | $3.96 | $1.98 | 155.87 | — | Peak rate. Off-peak is half. Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday, excluding Chinese public holidays. |
| 20 | GPT-6.1 Terra | OpenAIListed stakeMSFT | $1 | $5 | $2 | 168.11 | — | Short-context rate, for prompts up to 272K input tokens. Longer prompts are priced higher on the same page. |
| 21 | Muse Spark 1.3 | MetaListed companyMETA | $1.25 | $4.25 | $2 | 157.37 | — | Standard mode. Contributor mode is cheaper and lets Meta train on your content. That rate is not ranked here. |
| 22 | Grok 4.7 | xAIWholly owned by the listed companySPCX | $2 | $6 | $3 | 154.55 | — | Rate below 200,000 prompt tokens. At or above that, the whole request is billed at the long-context rate on the same page. |
| 23 | Qwen3.8 Max | AlibabaListed companyBABA9988.HK | $2 | $6 | $3 | 157.98 | 1483.3 | International price, Singapore region, on Alibaba Cloud Model Studio. Other regions on the same page differ. |
| 24 | Claude Sonnet 5.5 | AnthropicListed stakeGOOGLAMZN | $2 | $10 | $4 | 166.74 | — | Claude 4.7 and later use a newer tokenizer, about 30% more tokens for the same text. |
| 25 | Gemini 4 Argon | GoogleListed companyGOOGL | $2 | $10 | $4 | 166.07 | — | Standard paid rate through 30 November 2026, then $4 input and $20 output. Not on the free tier. Batch and Flex are half; Priority is higher. |
| 26 | GPT-6.1 Sol | OpenAIListed stakeMSFT | $2 | $10 | $4 | 167.82 | — | Short-context rate, for prompts up to 272K input tokens. Longer prompts are priced higher on the same page. |
| 27 | Claude Opus 5.5 | AnthropicListed stakeGOOGLAMZN | $4 | $20 | $8 | 168.4 | — | Claude 4.7 and later use a newer tokenizer, about 30% more tokens for the same text. |
| 28 | GPT-5.6 Sol | OpenAIListed stakeMSFT | $4 | $20 | $8 | 162.55 | — | Short-context promotional rate, for prompts up to 272K input tokens, in force at least through 21 November 2026. The page does not state the price after that. Longer prompts cost more on the same page. |
| 29 | Muse Wonder 1.0 | MetaListed companyMETA | $4 | $20 | $8 | 161.71 | — | Standard mode. Contributor mode is cheaper and lets Meta train on your content. That rate is not ranked here. |
| 30 | Claude Fable 5 | AnthropicListed stakeGOOGLAMZN | $10 | $50 | $20 | 163.02 | — | Claude 4.7 and later use a newer tokenizer, about 30% more tokens for the same text. |
| 31 | Claude Fable 5.1 | AnthropicListed stakeGOOGLAMZN | $10 | $50 | $20 | 166.36 | — | Claude 4.7 and later use a newer tokenizer, about 30% more tokens for the same text. |
| 32 | GPT-6 Astra | OpenAIListed stakeMSFT | $10 | $50 | $20 | 169.06 | — | Short-context rate, for prompts up to 272K input tokens. Longer prompts are priced higher on the same page. |
Input, output and blended are US dollars per million tokens.
Read this before you trust a rank
- The same model can cost a different amount on another host. These are the lab's own API, standard rate, read on 2 November 2026.
- Cached input and batch jobs are cheaper than the rates in this table. Those discounts are not in the rank.
- The blend assumes three input tokens for every one output token. A long answer costs more than this; a short one costs less.
- Some prices are promotions with an end date in the note. When that date passes, the rank moves.
- DeepSeek's off-peak rate is half the peak rate shown here. Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday, excluding Chinese public holidays.
- Claude 4.7 and later count about 30% more tokens for the same text than Claude Sonnet 4.6 and earlier. A lower price per token is not a lower price per page.
Sources
- OpenAI, read 2 November 2026. platform.openai.com
- Anthropic, read 2 November 2026. docs.anthropic.com
- Google, read 2 November 2026. ai.google.dev
- xAI, read 2 November 2026. docs.x.ai
- Mistral, read 2 November 2026. docs.mistral.ai
- DeepSeek, read 2 November 2026. api-docs.deepseek.com
- Alibaba, read 2 November 2026. alibabacloud.com
- Meta, read 2 November 2026. dev.meta.ai
Capability scores: the Epoch AI Capabilities Index, under the Creative Commons Attribution licence, estimates downloaded 2 November 2026. Epoch AI index · ECI estimates
Text ratings, only where the arena's model name matched ours exactly: the LMArena leaderboard dataset, Creative Commons Attribution 4.0, published 28 October 2026. A blank means we would not guess. LMArena dataset
Free · No card
See every AI tool — free
Join free as Spectator. The tools stay open for you after you sign in — same login as the rest of AllFactsAI.
- AI earnings calendar
- Stock screener
- Model leaderboard and prices
- AI capex tracker
- Company exposure cards
- AI value chain
- Daily and weekly Receipts
Free · No card · Unsubscribe anytime
Want the full picture? Insider
- Deep-dive company insights, built from public filings
- Capex leaderboard history and each company's spend over time
- Screener tools: 52-week range, personal watchlist, saved views
- Search the whole fact library and both newsletter archives
- Email alerts when a new fact lands on a company you follow
From US$9 a month · Opens to everyone Mon 23 Nov
See Insider →Already a member? Sign in