Score-per-dollar analysis across 300 AI models. Find the models that deliver the most performance for every dollar spent, broken down by price tier, provider, and the sweet spot between quality and affordability.
| # | Model | Score | Value Score |
|---|---|---|---|
| 1 | Llama 3.1 8B InstructMeta | 44 | 1251.4 |
| 2 | Qwen3 235B A22B Instruct 2507Alibaba | 65 | 756.7 |
| 3 | Qwen3.5-9BAlibaba | 67 | 700.0 |
| 4 | gpt-oss-20bOpenAI | 57 | 675.3 |
| 5 | Gemma 3 4BGoogle | 40 | 666.7 |
| 6 | Granite 4.0 MicroIBM | 40 | 629.9 |
| 7 | Mistral Small 3Mistral AI | 40 | 615.4 |
| 8 | Phi 4Microsoft | 60 | 587.3 |
| 9 | Granite 4.1 8BIBM | 40 | 533.3 |
| 10 | LFM2-24B-A2BLiquid AI | 40 | 533.3 |
| 11 | Qwen-TurboAlibaba | 40 | 492.3 |
| 12 | Gemma 3 12BGoogle | 40 | 470.6 |
| 13 | Nova Micro 1.0Amazon | 40 | 457.1 |
| 14 | Gemma 3n 4BGoogle | 40 | 444.4 |
| 15 | Hy3 previewTencent | 69 | 423.3 |
| 16 | Qwen3.5-FlashAlibaba | 69 | 422.8 |
| 17 | Trinity Miniarcee-ai | 40 | 410.3 |
| 18 | Reka Edgerekaai | 40 | 400.0 |
| 19 | Ministral 3 3B 2512Mistral AI | 40 | 400.0 |
| 20 | Nemotron Nano 9B V2NVIDIA | 40 | 400.0 |
| 21 | Gemma 4 26B A4B Google | 73 | 374.4 |
| 22 | gpt-oss-120bOpenAI | 41 | 369.9 |
| 23 | DeepSeek V4 FlashDeepSeek | 72 | 343.3 |
| 24 | Step 3.5 FlashStepFun | 67 | 334.0 |
| 25 | Gemma 3 27BGoogle | 40 | 333.3 |
| 26 | Nemotron 3 Nano 30B A3BNVIDIA | 40 | 320.0 |
| 27 | Llama 3.3 70B InstructMeta | 67 | 318.1 |
| 28 | Gemini 2.5 Flash Lite Preview 09-2025Google | 79 | 316.4 |
| 29 | Gemini 2.5 Flash LiteGoogle | 79 | 316.4 |
| 30 | Gemma 4 31BGoogle | 81 | 315.7 |
| 31 | Gemini 2.0 Flash LiteGoogle | 59 | 314.7 |
| 32 | Mistral Small 3.2 24BMistral AI | 40 | 290.9 |
| 33 | Gemini 2.0 FlashGoogle | 72 | 289.2 |
| 34 | Llama 4 ScoutMeta | 54 | 285.3 |
| 35 | GLM 4.7 FlashZhipu AI | 64 | 277.0 |
| 36 | Qwen3 8BAlibaba | 61 | 269.3 |
| 37 | Qwen3 30B A3B Thinking 2507Alibaba | 64 | 267.1 |
| 38 | Rnj 1 Instructessentialai | 40 | 266.7 |
| 39 | Ministral 3 8B 2512Mistral AI | 40 | 266.7 |
| 40 | UI-TARS 7B ByteDance | 40 | 266.7 |
| 41 | Qwen3 14BAlibaba | 40 | 266.7 |
| 42 | Reka Flash 3rekaai | 40 | 266.7 |
| 43 | Nova Lite 1.0Amazon | 40 | 266.7 |
| 44 | Ling-2.6-flashinclusionai | 40 | 250.0 |
| 45 | Phi 4 Mini InstructMicrosoft | 53 | 245.1 |
| 46 | Llama 3.3 Nemotron Super 49B V1.5NVIDIA | 61 | 242.4 |
| 47 | Devstral Small 1.1Mistral AI | 47 | 236.0 |
| 48 | Qwen3 30B A3BAlibaba | 64 | 235.9 |
| 49 | Qwen3 Coder 30B A3B InstructAlibaba | 40 | 235.3 |
| 50 | ERNIE 4.5 21B A3B ThinkingBaidu | 40 | 228.6 |
| # | Model | Score |
|---|---|---|
| 1 | Gemma 4 31B (free)Google | 81 |
| 2 | MiniMax M2.5 (free)MiniMax | 78 |
| 3 | Gemma 4 26B A4B (free)Google | 73 |
| 4 | GLM 4.5 Air (free)Zhipu AI | 71 |
| 5 | Qwen3 Next 80B A3B Instruct (free)Alibaba | 67 |
| 6 | Llama 3.3 70B Instruct (free)Meta | 66 |
| 7 | gpt-oss-20b (free)OpenAI | 57 |
| 8 | Ring-2.6-1T (free)inclusionai | 40 |
| 9 | CoBuddy (free)Baidu | 40 |
| 10 | Nemotron 3 Nano Omni (free)NVIDIA | 40 |
| 11 | Laguna XS.2 (free)poolside | 40 |
| 12 | Laguna M.1 (free)poolside | 40 |
| 13 | Qianfan-OCR-Fast (free)Baidu | 40 |
| 14 | Lyria 3 Pro PreviewGoogle | 40 |
| 15 | Lyria 3 Clip PreviewGoogle | 40 |
| 16 | Nemotron 3 Super (free)NVIDIA | 40 |
| 17 | LFM2.5-1.2B-Thinking (free)Liquid AI | 40 |
| 18 | LFM2.5-1.2B-Instruct (free)Liquid AI | 40 |
| 19 | Nemotron 3 Nano 30B A3B (free)NVIDIA | 40 |
| 20 | Nemotron Nano 12B 2 VL (free)NVIDIA | 40 |
| 21 | Nemotron Nano 9B V2 (free)NVIDIA | 40 |
| 22 | Qwen3 Coder 480B A35B (free)Alibaba | 40 |
Models in the top 20% for value score AND top 50% for composite score. The best of both worlds.
| Provider | Models | Avg Score | Avg Value |
|---|---|---|---|
| IBM | 2 | 40 | 581.6 |
| Microsoft | 2 | 56 | 416.2 |
| Meta | 8 | 54 | 333.4 |
| rekaai | 2 | 40 | 333.3 |
| NVIDIA | 4 | 45 | 277.6 |
| Tencent | 2 | 55 | 268.0 |
| 19 | 71 | 222.4 | |
| Amazon | 4 | 45 | 193.1 |
| arcee-ai | 6 | 48 | 178.2 |
| Mistral AI | 18 | 46 | 144.5 |
The value score is computed as composite quality score divided by average cost per million tokens. A model scoring 80 at $0.50/1M tokens has a value score of 160, while a model scoring 90 at $15/1M tokens has a value score of 6. Higher values mean more quality per dollar spent. Free models are excluded from value rankings since division by zero is undefined.
Sweet Spot models sit in the top 20% for value score AND the top 50% for composite quality score. They represent the best of both worlds: genuinely high-quality models that also deliver excellent value for money. These are the models most likely to satisfy both quality requirements and budget constraints.
Budget tier models (under $1/1M tokens) typically offer the highest value scores because even moderate quality at very low cost produces a strong ratio. However, the sweet spot analysis shows that some mid-tier models deliver the best combination of absolute quality and value. The ideal choice depends on whether you prioritize raw quality or cost efficiency.