Best Value LLM APIs
Ranked by value — Epoch Capabilities Index (ECI) ÷ blended $/1M tokens, i.e. capability delivered per dollar. Only models with both a published ECI and a published price qualify, and only those scoring at or above the qualifying set's median ECI (154.0, out of 86 qualifying models) make the cut — this keeps an ultra-cheap-but-weak model from gaming a pure price-per-point ranking.
| Rank | Model | Developer | ECI | Blended $/1M | ECI per $1 (blended, per 1M tokens) |
|---|---|---|---|---|---|
| 1 | Claude Haiku 5.5 | Anthropic | 160.2 | 0.20 | 801.1 |
| 2 | GPT-6 Luna | OpenAI | 156.7 | 0.20 | 783.4 |
| 3 | DeepSeek V4.1 Flash 1017 | DeepSeek | 159.3 | 0.2625 | 606.9 |
| 4 | GPT-5.6 Luna | OpenAI | 156.8 | 0.45 | 348.4 |
| 5 | DeepSeek V4 Pro 0813 | DeepSeek | 155.9 | 0.99 | 157.4 |
| 6 | Mistral Large 4 | Mistral | 154.7 | 1.05 | 147.3 |
| 7 | Gemini 3.8 Flash | 158.6 | 1.50 | 105.8 | |
| 8 | Gemini 3.7 Flash | 158.4 | 1.50 | 105.6 | |
| 9 | Gemini 3.6 Flash | 155.2 | 1.50 | 103.5 | |
| 10 | GPT-6.1 Terra | OpenAI | 167.8 | 2.00 | 83.9 |
| 11 | GLM-5.3 | Zhipu AI | 156.9 | 1.9421 | 80.8 |
| 12 | Muse Spark 1.3 | Meta | 158.0 | 2.00 | 79.0 |
| 13 | Muse Spark 1.2 | Meta | 155.1 | 2.00 | 77.5 |
| 14 | Muse Spark 1.1 | Meta | 154.9 | 2.00 | 77.4 |
| 15 | Grok 4.6 | xAI | 157.9 | 3.00 | 52.6 |
| 16 | Grok 4.7 | xAI | 156.1 | 3.00 | 52.0 |
| 17 | Qwen3.8 Max 0902 | Alibaba | 155.8 | 3.00 | 51.9 |
| 18 | Grok 4.5 | xAI | 155.1 | 3.00 | 51.7 |
| 19 | Gemini 3.5 Flash | 155.2 | 3.375 | 46.0 | |
| 20 | GPT-6.1 Sol | OpenAI | 168.9 | 4.00 | 42.2 |
| 21 | Gemini 4 Argon | 166.4 | 4.00 | 41.6 | |
| 22 | Claude Sonnet 5.5 | Anthropic | 166.2 | 4.00 | 41.6 |
| 23 | Qwen3.7 Max | Alibaba | 154.5 | 3.75 | 41.2 |
| 24 | GPT-6 Sol | OpenAI | 164.1 | 4.00 | 41.0 |
| 25 | Claude Sonnet 5 | Anthropic | 156.7 | 4.00 | 39.2 |
| 26 | GPT-5.6 Terra | OpenAI | 160.9 | 4.50 | 35.8 |
| 27 | GPT-5.2 | OpenAI | 154.4 | 4.8125 | 32.1 |
| 28 | GPT-5.4 | OpenAI | 157.8 | 5.625 | 28.1 |
| 29 | Kimi K3 | Moonshot AI | 158.5 | 6.00 | 26.4 |
| 30 | Claude Opus 5.5 | Anthropic | 168.4 | 8.00 | 21.0 |
| 31 | GPT-5.6 Sol | OpenAI | 163.0 | 8.00 | 20.4 |
| 32 | Muse Wonder | Meta | 162.0 | 8.00 | 20.3 |
| 33 | Claude Opus 5 | Anthropic | 164.4 | 10.00 | 16.4 |
| 34 | Claude Opus 4.8 | Anthropic | 159.9 | 10.00 | 16.0 |
| 35 | Claude Opus 4.7 | Anthropic | 157.6 | 10.00 | 15.8 |
| 36 | Claude Opus 4.6 | Anthropic | 156.3 | 10.00 | 15.6 |
| 37 | GPT-5.5 | OpenAI | 160.0 | 11.25 | 14.2 |
| 38 | GPT-6 Astra | OpenAI | 169.1 | 20.00 | 8.5 |
| 39 | Claude Fable 5.1 | Anthropic | 166.9 | 20.00 | 8.3 |
| 40 | Claude Fable 5 | Anthropic | 164.2 | 20.00 | 8.2 |
| 41 | GPT-5.2 Pro | OpenAI | 157.2 | 57.75 | 2.7 |
| 42 | GPT-5.5 Pro | OpenAI | 162.7 | 67.50 | 2.4 |
| 43 | GPT-5.4 Pro | OpenAI | 160.2 | 67.50 | 2.4 |
FAQ
How is 'value' calculated?
Value = Epoch Capabilities Index (ECI) ÷ blended $/1M tokens — capability points delivered per dollar spent. Higher is better.
Why is there an ECI floor?
Without one, an extremely weak but extremely cheap model would dominate a pure price-per-point ranking. Only models scoring at or above the median ECI (154.0) of the 86 models that have both a published ECI and a published price make this list.
What is the best-value LLM API right now?
Currently Claude Haiku 5.5 by Anthropic, delivering 801.1 ECI points per $1/1M blended tokens.
How is this different from the Cheapest APIs page?
/cheapest/ ranks purely by price, ignoring capability. This page ranks by capability delivered per dollar, so a slightly pricier but much stronger model can outrank a barely-cheaper, much weaker one.