Best Value AI Models
Ranked from provider-published pricing · Prices checked 12 August 2026
Value needs both halves of the equation. This list applies a capability floor first (a GPQA Diamond score of at least 60) and then ranks whatever clears it by blended cost per million tokens, cheapest first.
That is a deliberately transparent method rather than a composite score we invented: the bar is a published benchmark result, the ordering is a published price, and both are visible in the table.
The ranking
top 20 of 44Gemini 2.0 Flash from Google leads this ranking at $0.190/M. That is 95% below the $3.80/M median across the 44 models that qualify for this list.
| # | Model | Provider | Cheapest Models That Still Score Well | GPQA Diamond | Input /1M | Output /1M | Context |
|---|---|---|---|---|---|---|---|
| 1 | Gemini 2.0 Flash | $0.190/M | 62 | $0.100 | $0.400 | 1,000,000 | |
| 2 | Gemini 2.5 Flash | $0.285/M | 70 | $0.150 | $0.600 | 1,000,000 | |
| 3 | DeepSeek-R1-Distill-32B | DeepSeek | $0.290/M | 62 | $0.200 | $0.500 | 128,000 |
| 4 | Llama 4 Maverick | Meta | $0.320/M | 67 | $0.200 | $0.600 | 128,000 |
| 5 | Grok 3 mini | xAI | $0.360/M | 64 | $0.300 | $0.500 | 131,000 |
| 6 | GPT-5.6 Luna | OpenAI | $0.500/M | 92.3 | $0.200 | $1.20 | 1,050,000 |
| 7 | DeepSeek-R1-Distill-70B | DeepSeek | $0.509/M | 65 | $0.350 | $0.880 | 128,000 |
| 8 | DeepSeek-V3-0324 | DeepSeek | $0.519/M | 60 | $0.270 | $1.10 | 128,000 |
| 9 | DeepSeek-R1 (Cerebras) | Cerebras | $0.682/M | 71 | $0.550 | $0.990 | 64,000 |
| 10 | GPT-4.1 mini | OpenAI | $0.760/M | 60 | $0.400 | $1.60 | 1,000,000 |
| 11 | DeepSeek-R1 (Groq) | Groq | $0.822/M | 71 | $0.750 | $0.990 | 128,000 |
| 12 | DeepSeek-R1 | DeepSeek | $1.04/M | 71 | $0.550 | $2.19 | 64,000 |
| 13 | DeepSeek-R1-0528 | DeepSeek | $1.04/M | 73 | $0.550 | $2.19 | 64,000 |
| 14 | DeepSeek-R1 (DI) | DeepInfra | $1.04/M | 71 | $0.550 | $2.19 | 64,000 |
| 15 | QwQ-32B | Qwen | $1.14/M | 65 | $0.600 | $2.40 | 131,000 |
| 16 | Llama-3.1-Nemotron-Ultra-253B | Nvidia | $1.60/M | 60 | $1.60 | $1.60 | 128,000 |
| 17 | o3-mini | OpenAI | $2.09/M | 79 | $1.10 | $4.40 | 200,000 |
| 18 | o4-mini | OpenAI | $2.09/M | 81 | $1.10 | $4.40 | 200,000 |
| 19 | Sonar Reasoning | Perplexity | $2.20/M | 60 | $1.00 | $5.00 | 127,000 |
| 20 | Qwen2.5-Max | Qwen | $3.04/M | 60 | $1.60 | $6.40 | 32,000 |
How this list is built
- Ranked cheapest first by a blended rate weighting input and output at 70/30. Default reference rate. 70% weight on input, 30% on output.
- Excludes embed models.
- Requires a GPQA Diamond score of at least 60.
- Includes models marked available or preview; retired and deprecated SKUs are excluded.
- Prices are the rates each provider publishes, not estimates. How the blended rate is calculated.
Frequently asked questions
How is "best value" decided here?
Two published numbers, in order. First we keep only models scoring at least 60 on GPQA Diamond, a graduate-level reasoning benchmark. Then we rank the survivors by blended cost per million tokens, cheapest first. There is no weighting or composite index, just a capability floor and a price sort, both shown in the table.
Why a GPQA score of 60?
It is a round threshold that separates models capable of genuine multi-step reasoning from those that mostly pattern-match, without being so high that only the frontier tier survives. Models we hold no GPQA score for are excluded rather than assumed to pass, the count of those exclusions is shown beneath the table.
Related rankings
Price it for your workload
Rankings are recomputed on every deploy from the catalogue. Prices checked 12 August 2026.