Model Rankings
Every model on Frontière AI is ranked by two criteria: Best Value (capability-to-price ratio — bigger, smarter models at lower prices rank higher) and Best Performance (raw capability, verified score, multimodality, and context length). Sovereign EU models get a +30% boost in the value ranking.
Rankings
20B · 6/6 ✓ · EU sovereign · €0.0056/100k in
30.5B · 6/6 ✓ · EU sovereign · €0.0084/100k in
120B · 6/6 ✓ · EU sovereign · €0.0112/100k in
552B · 6/6 ✓ · Fast access · €0.0182/100k in
32.8B · 6/6 ✓ · EU sovereign · €0.0112/100k in
552B · 6/6 ✓ · Fast access · €0.0182/100k in
2.8T · 5/6 ✓ · Fast access · €0.3654/100k in
753B · 6/6 ✓ · EU sovereign · €0.252/100k in
403B · 6/6 ✓ · EU sovereign · €0.084/100k in
158B · 6/6 ✓ · EU sovereign · €0.056/100k in
| # | Model | Size | Price (input) | Jurisdiction | Verified Score |
|---|---|---|---|---|---|
| 1 | GPT-OSS 20B (OpenAI) | 20B | €0.0056 | EU sovereign | 6/6 |
| 2 | Qwen3 Coder 30B | 30.5B | €0.0084 | EU sovereign | 6/6 |
| 3 | GPT-OSS 120B (OpenAI) | 120B | €0.0112 | EU sovereign | 6/6 |
| 4 | DeepSeek V4.1 Flash (552B MoE, 1M context) | 552B | €0.0182 | Fast access | 6/6 |
| 5 | Qwen3 32B | 32.8B | €0.0112 | EU sovereign | 6/6 |
| 6 | Mistral Small 3.2 24B | 24B | €0.0126 | EU sovereign | 6/6 |
| 7 | Gemma 3 27B (Google) | 27B | €0.0098 | Fast access | 0/6 |
| 8 | Qwen3.5 9B (fast, budget) | 9.65B | €0.014 | EU sovereign | 6/6 |
| 9 | DeepSeek R1 (Distill Qwen 14B) | 14B | €0.0098 | Fast access | 0/6 |
| 10 | DeepSeek R1 (Distill Qwen 32B) | 32B | €0.0154 | Fast access | 0/6 |
| 11 | DeepSeek V4 Flash 0731 (1M context) | 158B | €0.056 | EU sovereign | 6/6 |
| 12 | Qwen3.5 397B (multimodal) | 403B | €0.084 | EU sovereign | 6/6 |
| 13 | Qwen3.8 27B (multimodal) | 27.8B | €0.056 | EU sovereign | 5/6 |
| 14 | Qwen3.6 27B (multimodal) | 27.8B | €0.0574 | EU sovereign | 6/6 |
| 15 | DeepSeek R1 (671B MoE) | 671B | €0.0854 | Fast access | 0/6 |
| 16 | Llama 3.3 70B | 70B | €0.0898 | EU sovereign | 5/6 |
| 17 | Qwen3 235B (instruct) | 235B | €0.105 | EU sovereign | 0/6 |
| 18 | Qwen2.5-VL 72B (vision) | 73.4B | €0.1232 | EU sovereign | 5/6 |
| 19 | DeepSeek V4 Pro (345B MoE) | 345B | €0.1064 | Fast access | 0/6 |
| 20 | GLM-5.2 (Zhipu/Z.ai) | 753B | €0.252 | EU sovereign | 6/6 |
| 21 | Qwen3.8 Max (2.4T, 1M context) | 2.4T | €0.2422 | Fast access | 0/6 |
| 22 | Kimi K3 (2.8T parameters, 1M context) | 2.8T | €0.3654 | Fast access | 5/6 |
| 23 | Muse Spark 1.1 (1M context) | ? | €0.1512 | Fast access | 0/6 |
| 24 | Muse Spark 1.2 (1M context) | ? | €0.1512 | Fast access | 0/6 |
| 25 | GLM-5.3 Flash (Zhipu/Z.ai) | ? | €0.0098 | Fast access | 0/6 |
| 26 | Qwen3.8 Flash (Qwen) | ? | €0.0182 | Fast access | 0/6 |
| 27 | GLM-5.3 (Zhipu/Z.ai) | ? | €0.1694 | Fast access | 0/6 |
Two rankings: Best Value computes a capability-to-price score (log-scale parameters × sovereignty +30% × context bonus × multimodality × verified score) divided by input price. Best Performance ranks by raw capability (parameters × verified criteria × features). The goal: help you pick the right model for your use case, not just the cheapest one.