Best LLM APIs by intelligence per dollar
A second take on value, from a different benchmark. This ranks56 models by the Artificial Analysis Intelligence Index — an aggregate 0–100 intelligence score — divided by each model's cheapest blended price per 1M tokens. Where the Elo value boardmeasures human preference, this measures aggregated reasoning and knowledge. Each model links to its cheapest provider; all figures computed from the index.
Ranked by intelligence per dollar
| # | Model | Index | Cheapest /1M | Verified | Index / $ | |
|---|---|---|---|---|---|---|
| 1 | deepseek-v4-flash · Baidu | 51.8 | $0.15 | today | 345.3 | |
| 2 | glm-5.3-flash · Relace | 57.5 | $0.309 | 1d ago | 186.2 | |
| 3 | nemotron-3.5-lightning · DeepInfra | 23.6 | $0.28 | today | 84.3 | |
| 4 | granite-4.2-8b · CoreWeave | 19.6 | $0.25 | today | 78.4 | |
| 5 | gpt-5.6-luna · OpenAI | 51.2 | $0.7 | today | 73.1 | |
| 6 | gemma-4-31b · CoreWeave | 29.7 | $0.44 | today | 67.5 | |
| 7 | gemma-4-26b-a4b · Cloudflare | 26.1 | $0.4 | today | 65.2 | |
| 8 | nemotron-3-super-120b-a12b · DeepInfra | 25.7 | $0.485 | today | 53 | |
| 9 | glm-4.7-flash · Venice | 23.3 | $0.46 | today | 50.7 | |
| 10 | deepseek-v3.2 · GMICloud | 25.1 | $0.518 | today | 48.4 | |
| 11 | minimax-m3 · CoreWeave | 45.4 | $1.19 | today | 38.2 | |
| 12 | ministral-3b · Mistral AI | 7.1 | $0.2 | today | 35.5 | |
| 13 | qwen3.6-35b-a3b · AkashML | 32.1 | $1 | today | 32.1 | |
| 14 | ministral-8b · Mistral AI | 9 | $0.3 | today | 30 | |
| 15 | minimax-m2.7 · Novita AI | 38.9 | $1.35 | today | 28.8 | |
| 16 | minimax-m2.5 · Venice | 34.5 | $1.22 | today | 28.3 | |
| 17 | ministral-14b · Mistral AI | 11.2 | $0.4 | today | 28 | |
| 18 | qwen3.5-35b-a3b · DeepInfra | 29.9 | $1.14 | today | 26.2 | |
| 19 | qwen3.8-27b · Parasail | 52 | $2.44 | today | 21.3 | |
| 20 | qwen3.5-27b · Alibaba | 34.6 | $1.755 | today | 19.7 | |
| 21 | qwen3.6-27b · Chutes | 37.7 | $2.3 | today | 16.4 | |
| 22 | glm-5 · GMICloud | 40.5 | $2.52 | today | 16.1 | |
| 23 | glm-4.7 · DeepInfra | 34.5 | $2.15 | today | 16 | |
| 24 | kimi-k2.6 · Baidu | 45.1 | $2.856 | today | 15.8 | |
| 25 | nemotron-3-ultra-550b-a55b · DeepInfra | 38.3 | $2.7 | today | 14.2 | |
| 26 | qwen3.5-122b-a10b · SiliconFlow | 32.9 | $2.34 | today | 14.1 | |
| 27 | kimi-k2.5 · SiliconFlow | 36 | $2.7 | today | 13.3 | |
| 28 | gemini-3.8-flash · Google | 58.7 | $4.5 | 2d ago | 13 | |
| 29 | glm-5.3 · Reka | 59.5 | $4.65 | today | 12.8 | |
| 30 | qwen3.5-397b-a17b · Alibaba | 34.3 | $2.73 | today | 12.6 | |
| 31 | gemini-3.6-flash · Google | 51.6 | $4.5 | 2d ago | 11.5 | |
| 32 | glm-5.1 · Baidu | 41 | $3.764 | today | 10.9 | |
| 33 | glm-4.6 · Venice | 23.4 | $2.18 | today | 10.7 | |
| 34 | kimi-k2.7-code · DeepInfra | 43 | $4.08 | today | 10.5 | |
| 35 | gpt-5.6-sol · OpenAI | 58.9 | $6 | today | 9.8 | |
| 36 | gpt-5.6-terra · OpenAI | 55 | $7 | today | 7.9 | |
| 37 | grok-4.6 · xAI | 60.9 | $8 | 2d ago | 7.6 | |
| 38 | claude-sonnet-5 · Anthropic | 55.3 | $12 | today | 4.6 | |
| 39 | kimi-k3 · DeepInfra | 59.7 | $17.1 | today | 3.5 | |
| 40 | mistral-medium-3-5 · Mistral AI | 30.4 | $9 | today | 3.4 | |
| 41 | claude-opus-5 · Anthropic | 63 | $30 | today | 2.1 | |
| 42 | claude-fable-5 · Anthropic | 62.1 | $60 | today | 1 |
Value = index score ÷ (input + output price per 1M tokens). A model's score is identical across providers, so the cheapest host gives the best value — that's the price shown. Models released within the last 12 months; older ones below. Blended is a coarse ranking aid, not a workload estimate; size your own on each model's page.
Older models — released more than 12 months ago (14)
Still priced daily and often still the pragmatic choice, but no longer the current generation.
| # | Model | Index | Cheapest /1M | Age | Index / $ |
|---|---|---|---|---|---|
| 43 | gpt-oss-20b · AkashML | 15.2 | $0.12 | 13 mo | 126.7 |
| 44 | gpt-oss-120b · AkashML | 24.1 | $0.2 | 13 mo | 120.5 |
| 45 | gemma-3-27b · DeepInfra | 7.4 | $0.24 | 17 mo | 30.8 |
| 46 | llama-4-scout · DeepInfra | 10.3 | $0.4 | 17 mo | 25.8 |
| 47 | deepseek-chat-v3.1 · DeepInfra | 21.4 | $1.2 | 12 mo | 17.8 |
| 48 | glm-4.5-air · Novita AI | 16.7 | $0.98 | 13 mo | 17 |
| 49 | llama-4-maverick · DigitalOcean | 14.5 | $0.896 | 17 mo | 16.2 |
| 50 | deepseek-v3 · DeepInfra | 14.2 | $1.21 | 20 mo | 11.7 |
| 51 | gpt-4o-mini · OpenAI | 6.7 | $0.75 | 25 mo | 8.9 |
| 52 | deepseek-r1 · DeepInfra | 20.4 | $2.65 | 15 mo | 7.7 |
| 53 | gpt-4.1-mini · OpenAI | 14.8 | $2 | 16 mo | 7.4 |
| 54 | minimax-m1 · Minimax | 17.9 | $2.6 | 14 mo | 6.9 |
| 55 | mistral-medium-3 · Mistral AI | 12.5 | $2.4 | 15 mo | 5.2 |
| 56 | gpt-4o · OpenAI | 11.1 | $12.5 | 27 mo | 0.9 |
For contrast: the top of the intelligence index
The highest-scoring models by raw index — rarely the best value, which is the point.
| Index rank | Model | Index | Cheapest /1M | Index / $ |
|---|---|---|---|---|
| 1 | claude-opus-5 | 63 | $30 | 2.1 |
| 2 | claude-fable-5 | 62.1 | $60 | 1 |
| 3 | grok-4.6 | 60.9 | $8 | 7.6 |
| 4 | kimi-k3 | 59.7 | $17.1 | 3.5 |
| 5 | glm-5.3 | 59.5 | $4.65 | 12.8 |
| 6 | gpt-5.6-sol | 58.9 | $6 | 9.8 |
| 7 | gemini-3.8-flash | 58.7 | $4.5 | 13 |
| 8 | glm-5.3-flash | 57.5 | $0.309 | 186.2 |
| 9 | claude-sonnet-5 | 55.3 | $12 | 4.6 |
| 10 | gpt-5.6-terra | 55 | $7 | 7.9 |
Artificial Analysis Intelligence Index · verified 2026-09-05 · leaderboard source · full methodology. Two benchmarks disagree by design — cross-check the Elo value ranking.
FAQ
Which LLM API gives the most intelligence per dollar?
As of 2026-09-05, deepseek-v4-flash leads: 345.3 index points per dollar — a 51.8 Artificial Analysis Intelligence Index score at $0.15 per 1M tokens (blended input + output) via Baidu.
How is this different from the Elo value ranking?
This ranks by the Artificial Analysis Intelligence Index — an aggregate 0–100 reasoning and knowledge score — instead of LMArena Elo, which measures human preference. Two independent benchmarks cross-check each other; they disagree by design, so a model strong on one may rank differently on the other.
Is the smartest model the best value?
Rarely. The top model by raw index is claude-opus-5 at 63, but at $30 per 1M its value is 2.1 points per dollar — below the value leader deepseek-v4-flash. Peak intelligence usually carries a price premium.