Price comparison
First-party API list prices in USD per million tokens. Blended price weights input to output 3:1, close to the real mix in most chat and RAG workloads.
Models sold only by subscription, with no token-priced API (such as Muse Wonder Full), stay out of this table — they still appear on the leaderboard.
| Model | Input $/1M | Output $/1M | Blended | Cache $ | Intelligence v5 | pts/$ | Notes |
|---|---|---|---|---|---|---|---|
| gpt-oss-120b Open weightsOpenAI | $0.05 | $0.20 | $0.09 | — | — | — | |
| Qwen3.8 27B Open weightsAlibaba Qwen | $0.10 | $0.40 | $0.17 | — | 33.7 | — | |
| Gemma 4 31B Open weightsGoogle | $0.10 | $0.40 | $0.17 | — | 19.0 | — | |
| Claude Haiku 5.5Anthropic | $0.10 | $0.50 | $0.20 | $0.01 | 36.0 | 180 | |
| GPT-6 LunaOpenAI | $0.10 | $0.50 | $0.20 | $0.01 | — | — | |
| GLM-5.3 Flash Open weightsZ.ai | $0.15 | $0.50 | $0.24 | — | 41.8 | — | |
| MiniMax M3.1 FlashMiniMax | $0.15 | $0.60 | $0.26 | — | 29.0 | 111 | |
| DeepSeek V4.1 Flash Open weightsDeepSeek | $0.30 | $1.20 | $0.53 | $0.006 | 33.0 | 63 | Half price off-peak |
| MiniMax M3 Open weightsMiniMax | $0.30 | $1.20 | $0.53 | — | 29.2 | — | |
| MiMo-V2.6 Pro Open weightsXiaomi | $0.35 | $1.40 | $0.61 | — | 30.0 | 49 | |
| Nemotron 3 Ultra Open weightsNVIDIA | $0.40 | $1.60 | $0.70 | — | — | — | |
| Step 5 PreviewStepFun | $0.50 | $2.00 | $0.88 | — | — | — | |
| Mistral Large 4 Open weightsMistral | $0.70 | $2.10 | $1.05 | $0.07 | 26.0 | 25 | |
| Gemini 3.8 FlashGoogle | $0.75 | $3.75 | $1.50 | — | — | — | Doubles 01-01 |
| DeepSeek V4 Pro Open weightsDeepSeek | $1.32 | $3.96 | $1.98 | $0.04 | 23.0 | 12 | |
| GPT-6.1 TerraOpenAI | $1.00 | $5.00 | $2.00 | $0.05 | 51.3 | 26 | |
| GLM-5.3 Open weightsZ.ai | $1.40 | $4.40 | $2.15 | — | 32.0 | 15 | |
| Grok 4.7xAI | $2.00 | $6.00 | $3.00 | — | 34.3 | 11 | |
| Qwen3.8 MaxAlibaba Qwen | $2.00 | $6.00 | $3.00 | — | 34.0 | 11 | |
| Claude Sonnet 5.5Anthropic | $2.00 | $10.00 | $4.00 | $0.10 | 54.5 | 14 | |
| GPT-6.1 SolOpenAI | $2.00 | $10.00 | $4.00 | $0.10 | 54.6 | 14 | |
| Gemini 4 ArgonGoogle | $2.00 | $10.00 | $4.00 | — | 50.4 | 13 | Doubles 12-01 |
| Kimi K3.1 Pro Open weightsMoonshot AI | $2.00 | $10.00 | $4.00 | $0.20 | 46.0 | 12 | |
| Kimi K3 Open weightsMoonshot AI | $3.00 | $15.00 | $6.00 | — | 43.6 | — | |
| Claude Opus 5.5Anthropic | $4.00 | $20.00 | $8.00 | $0.20 | 59.1 | 7 | |
| Gemini 4 CarbonGoogle | $4.00 | $20.00 | $8.00 | $0.20 | 60.0 | 8 | |
| Muse WonderMeta | $4.00 | $20.00 | $8.00 | — | 44.0 | 6 | |
| Claude Fable 5.1Anthropic | $10.00 | $50.00 | $20.00 | $0.25 | 49.0 | 2 | |
| GPT-6 AstraOpenAI | $10.00 | $50.00 | $20.00 | $1.00 | 56.9 | 3 |
In the Intelligence v5 column a grey number is the v4.3.2 score used while the model has no v5 result (a looser scale — do not compare it with black figures), and an italic dash means neither scale covers it.
Price × intelligence
Top-left is the ideal zone — cheap and smart. Bottom-right is the zone to avoid — expensive and mediocre. Bigger dots score higher.
By value
How many intelligence points one dollar buys. Only models with both a v5 score and a public price are counted.
| # | Model | pts/$ | v5 | Blended |
|---|---|---|---|---|
| 1 | Claude Haiku 5.5Anthropic | 180.0 | 36.0 | $0.20 |
| 2 | MiniMax M3.1 FlashMiniMax | 110.7 | 29.0 | $0.26 |
| 3 | DeepSeek V4.1 Flash Open weightsDeepSeek | 62.9 | 33.0 | $0.53 |
| 4 | MiMo-V2.6 Pro Open weightsXiaomi | 49.0 | 30.0 | $0.61 |
| 5 | GPT-6.1 TerraOpenAI | 25.6 | 51.3 | $2.00 |
| 6 | Mistral Large 4 Open weightsMistral | 24.8 | 26.0 | $1.05 |
| 7 | GLM-5.3 Open weightsZ.ai | 14.9 | 32.0 | $2.15 |
| 8 | GPT-6.1 SolOpenAI | 13.7 | 54.6 | $4.00 |
| 9 | Claude Sonnet 5.5Anthropic | 13.6 | 54.5 | $4.00 |
| 10 | Gemini 4 ArgonGoogle | 12.6 | 50.4 | $4.00 |
| 11 | DeepSeek V4 Pro Open weightsDeepSeek | 11.6 | 23.0 | $1.98 |
| 12 | Kimi K3.1 Pro Open weightsMoonshot AI | 11.5 | 46.0 | $4.00 |
Cheap is not the same as economical.DeepSeek V4.1 Flash buys more than 8× the intelligence points per dollar that Claude Opus 5.5 does. But if it needs 3× the output tokens for the same task, the real bill gap narrows to 2-3×.
Promo prices distort the picture.Gemini 4 Argon is $2/$10 today, level with Sonnet 5.5. When it doubles on 1 December its value ranking drops to the middle of the table. Any budget reaching beyond a month should use the restored price.
Cache pricing is the line everyone ignores.DeepSeek V4.1 Flash charges $0.003 on cache hits — 1% of its standard input price. Once your system prompt passes 2,000 tokens, caching strategy decides the bill more than model choice does.