Open-source AI models: quality and price
Instead of a subscription: download the weights and run them, or pay a provider by the token. Quality is the rank in independent blind tests; price is what providers charge per million tokens.
Open weights means exactly that: the creator's official Hugging Face organisation has published them. Quality and price from Artificial Analysis, downloads and license from Hugging Face. Collected 2026-10-26.
Language models
By intelligence index. Price is the blended rate per million tokens at 3 input : 1 output.
| Model | Quality | License | API price | Downloads / month |
|---|---|---|---|---|
| MiMo-V2.6-ProXiaomi · XiaomiMiMo/MiMo-V2.6-Pro-RL | #34 of 698intelligence index 46.3 | mit | $0.54₩700 per 1M tokens | 32,525 |
| Qwen3.8 Max (0902)Alibaba · Qwen/Qwen3.8-Flash-Next | #36 of 698intelligence index 45.4 | other | $3₩4,000 per 1M tokens | 3,093,369 |
| DeepSeek V4.1 Flash 1017 (Max)DeepSeek · deepseek-ai/DeepSeek-V4.1-Flash-1017 | #37 of 698intelligence index 45.1 | mit | $0.53₩700 per 1M tokens | 17,924 |
| GLM-5.3 (Max)Z AI · zai-org/GLM-5.3-Flash | #40 of 698intelligence index 44.8 | mit | $2.15₩2,800 per 1M tokens | 6,352,272 |
| Kimi K3 (Max)Kimi · moonshotai/Kimi-K3 | #47 of 698intelligence index 43.6 | other | $6₩7,900 per 1M tokens | 874,512 |
| Ling 3.1 FlashInclusionAI · inclusionAI/Ling-3.1-flash | #60 of 698intelligence index 41.1 | mit | $0.45₩600 per 1M tokens | 4,222 |
| Qwen3.8 2.4T A95BAlibaba · Qwen/Qwen3.8-2.4T-A95B | #65 of 698intelligence index 39.9 | other | $3₩4,000 per 1M tokens | 25,164 |
| DeepSeek V4.1 Flash (Max)DeepSeek · deepseek-ai/DeepSeek-V4.1-Flash | #71 of 698intelligence index 39.5 | mit | $0.52₩700 per 1M tokens | 1,640,989 |
| MiMo-V2.6-FlashXiaomi · XiaomiMiMo/MiMo-V2.6-Flash-RL | #83 of 698intelligence index 37.9 | mit | $0.18₩200 per 1M tokens | 89,572 |
| DeepSeek V4 Pro 0813 (Max)DeepSeek · deepseek-ai/DeepSeek-V4-Pro-0813 | #88 of 698intelligence index 36 | mit | $1.98₩2,600 per 1M tokens | 150,272 |
| DeepSeek V4 Flash Vision (Max)DeepSeek · deepseek-ai/DeepSeek-V4-Flash-Vision-Exp | #92 of 698intelligence index 34.8 | mit | $0.66₩900 per 1M tokens | 326,391 |
| DeepSeek V4 Flash 0731 (Max)DeepSeek · deepseek-ai/DeepSeek-V4-Flash-0731 | #97 of 698intelligence index 34.3 | mit | $0.18₩200 per 1M tokens | 4,112,191 |
| GLM-5.2 (Max)Z AI · zai-org/GLM-5.2 | #105 of 698intelligence index 33.7 | mit | $2.15₩2,800 per 1M tokens | 298,589 |
| Qwen3.8 27B (Xhigh)Alibaba · Qwen/Qwen3.8-27B | #106 of 698intelligence index 33.7 | apache-2.0 | $1.13₩1,500 per 1M tokens | 7,762,014 |
| MiniMax-M3MiniMax · MiniMaxAI/MiniMax-M3 | #133 of 698intelligence index 29.2 | other | $0.53₩700 per 1M tokens | 201,055 |
| GLM-5 (Reasoning)Z AI · zai-org/GLM-5 | #143 of 698intelligence index 27.9 | mit | $1.55₩2,100 per 1M tokens | 1,169,522 |
| Kimi K2.6 (Reasoning)Kimi · moonshotai/Kimi-K2.6 | #150 of 698intelligence index 27 | other | $1.71₩2,300 per 1M tokens | 353,428 |
| Solar Pro 4Upstage · upstage/solar-pro-preview-instruct | #153 of 698intelligence index 26.6 | mit | $0.53₩700 per 1M tokens | 4,949 |
| GLM-5.1 (Reasoning)Z AI · zai-org/GLM-5.1 | #160 of 698intelligence index 26.1 | mit | $1.98₩2,600 per 1M tokens | 726,699 |
| MiMo-V2.5-Pro (Reasoning)Xiaomi · XiaomiMiMo/MiMo-V2.5-Pro | #162 of 698intelligence index 26 | mit | $0.56₩700 per 1M tokens | 12,500 |
| Kimi K2.7 CodeKimi · moonshotai/Kimi-K2.7-Code | #163 of 698intelligence index 25.8 | other | $1.71₩2,300 per 1M tokens | 89,154 |
| Hy3Tencent · tencent/Hy3 | #167 of 698intelligence index 25.3 | apache-2.0 | $0.24₩300 per 1M tokens | 965,991 |
| Solar Open2 250BUpstage · upstage/Solar-Open2-250B | #176 of 698intelligence index 24.7 | other | $0₩0 per 1M tokens | 7,913 |
| Ling-3.0-flash-VLInclusionAI · inclusionAI/Ling-3.0-flash-VL | #180 of 698intelligence index 24.6 | mit | $0₩0 per 1M tokens | 12,161 |
| Kimi K2.5 (Reasoning)Kimi · moonshotai/Kimi-K2.5 | #194 of 698intelligence index 23.5 | other | $1.2₩1,600 per 1M tokens | 340,246 |
| Nemotron 3 Ultra 550B A55B (Reasoning)NVIDIA · nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-GenRM | #199 of 698intelligence index 22.9 | other | $1.13₩1,500 per 1M tokens | 808 |
| Qwen3.5 27B (Reasoning)Alibaba · Qwen/Qwen3.5-27B | #200 of 698intelligence index 22.9 | apache-2.0 | $0.83₩1,100 per 1M tokens | 1,734,331 |
| MiniMax-M2.5MiniMax · MiniMaxAI/MiniMax-M2.5 | #203 of 698intelligence index 22.8 | other | $0.53₩700 per 1M tokens | 1,019,217 |
| MiniMax-M2.7MiniMax · MiniMaxAI/MiniMax-M2.7 | #204 of 698intelligence index 22.8 | other | $0.53₩700 per 1M tokens | 422,497 |
| Ling-3.0-flash-FinInclusionAI · inclusionAI/Ling-3.0-flash-Fin | #207 of 698intelligence index 22.6 | mit | $0₩0 per 1M tokens | 1,780 |
148 more; the top 30 are shown.
Image generation
By Image Arena preference Elo.
| Model | Quality | License | Downloads / month |
|---|---|---|---|
| FLUX 3 ImageBlack Forest Labs · black-forest-labs/flux-3-image-klein-comfy | #13 of 106preference Elo 1,117 | other | 660,874 |
| Qwen-Image-2.1Alibaba · Qwen/Qwen-Image-2.1 | #22 of 106preference Elo 1,042 | other | 184,227 |
| Qwen-Image-2.1-TurboAlibaba · Qwen/Qwen-Image-2.1-Turbo | #27 of 106preference Elo 1,029 | other | 31,396 |
| FLUX.2 [max]Black Forest Labs · black-forest-labs/FLUX.2-dev | #35 of 106preference Elo 1,021 | other | 326,662 |
| Ming-Image-0.1-DesignInclusionAI · inclusionAI/Ming-Image-0.1-Design | #47 of 106preference Elo 999 | mit | 1,350 |
| Cosmos3-Super-Text2Image (agentic)NVIDIA · nvidia/Cosmos3-Super-Text2Image | #50 of 106preference Elo 993 | other | 7,764 |
| HunyuanImage 3.0 (Tencent Cloud)Tencent · tencent/HunyuanImage-3.0-Instruct | #54 of 106preference Elo 989 | other | 40,270 |
| HiDream-O1-ImageHiDream · HiDream-ai/HiDream-O1-Image | #63 of 106preference Elo 975 | mit | 4,514 |
| Cosmos3-Super-Text2Image-4StepNVIDIA · nvidia/Cosmos3-Super-Text2Image-4Step | #65 of 106preference Elo 970 | other | 465 |
| FLUX.2 [klein] 9BBlack Forest Labs · black-forest-labs/FLUX.2-klein-9B | #75 of 106preference Elo 929 | other | 188,764 |
| FLUX.1 Kontext [max]Black Forest Labs · black-forest-labs/FLUX.1-Kontext-dev | #76 of 106preference Elo 929 | other | 336,334 |
| ERNIE Image TurboBaidu · baidu/ERNIE-Image-Turbo | #79 of 106preference Elo 914 | apache-2.0 | 2,846 |
| Qwen ImageAlibaba · Qwen/Qwen-Image | #83 of 106preference Elo 903 | apache-2.0 | 307,477 |
| FLUX.1 Krea [dev]Black Forest Labs · black-forest-labs/FLUX.1-Krea-dev | #85 of 106preference Elo 896 | other | 32,879 |
| GLM-ImageZ AI · zai-org/GLM-Image | #88 of 106preference Elo 885 | mit | 4,997 |
| FLUX.2 [klein] 4BBlack Forest Labs · black-forest-labs/FLUX.2-klein-base-4B | #96 of 106preference Elo 837 | apache-2.0 | 562,486 |
| SRPOTencent · tencent/SRPO | #99 of 106preference Elo 826 | other | 2,028 |
| FLUX.1 [dev]Black Forest Labs · black-forest-labs/FLUX.1-dev | #100 of 106preference Elo 823 | other | 648,089 |
| HiDream-O1-Image-DevHiDream · HiDream-ai/HiDream-O1-Image-Dev | #102 of 106preference Elo 810 | mit | 590 |
| FLUX.1 [schnell]Black Forest Labs · black-forest-labs/FLUX.1-schnell | #104 of 106preference Elo 740 | apache-2.0 | 507,663 |
Video generation
By Video Arena preference Elo.
| Model | Quality | License | Downloads / month |
|---|---|---|---|
| MiniMax H3MiniMax · MiniMaxAI/MiniMax-H3 | #4 of 75preference Elo 1,356 | other | 5,173,270 |
| Cosmos3-Super-Image2Video-4StepNVIDIA · nvidia/Cosmos3-Super-Image2Video-4Step | #19 of 75preference Elo 1,275 | other | 619 |
| Cosmos3-Super-Image2VideoNVIDIA · nvidia/Cosmos3-Super-Image2Video | #30 of 75preference Elo 1,253 | other | 45,576 |
| LTX-2.5 FastLightricks · Lightricks/LTX-2.5 | #42 of 75preference Elo 1,215 | other | 2,107,108 |
| LTX-2 FastLightricks · Lightricks/LTX-2 | #48 of 75preference Elo 1,191 | other | 287,028 |
| LTX-2.3 FastLightricks · Lightricks/LTX-2.3 | #56 of 75preference Elo 1,150 | other | 831,585 |
| HunyuanVideo-1.5 (Fal)Tencent · tencent/HunyuanVideo-1.5 | #59 of 75preference Elo 1,134 | other | 1,352 |
| Wan 2.2 A14BAlibaba · Wan-AI/Wan2.2-T2V-A14B-Diffusers | #62 of 75preference Elo 1,107 | apache-2.0 | 170,405 |
| Wan 2.1 14BAlibaba · Wan-AI/Wan2.1-VACE-14B | #73 of 75preference Elo 1,000 | apache-2.0 | 122,916 |
| Wan 2.2 5BAlibaba · Wan-AI/Wan2.2-TI2V-5B-Diffusers | #74 of 75preference Elo 992 | apache-2.0 | 165,990 |
| Hunyuan Video (Fal)Tencent · tencent/HunyuanVideo | #75 of 75preference Elo 920 | other | 1,516 |
Speech synthesis
By Voice Arena preference Elo.
| Model | Quality | License | Downloads / month |
|---|---|---|---|
| Step Audio EditX (Mar 2026)StepFun · stepfun-ai/Step-Audio-EditX | #41 of 98preference Elo 1,100 | - | 2,536 |
| Magpie-Multilingual 357M (Feb 2026)NVIDIA · nvidia/magpie_tts_multilingual_357m | #59 of 98preference Elo 1,063 | other | 11,147 |
| VibeVoice 1.5BMicrosoft · microsoft/VibeVoice-1.5B | #83 of 98preference Elo 956 | mit | 324,867 |
| XTTS v2Coqui · coqui/XTTS-v2 | #90 of 98preference Elo 914 | other | 6,139,889 |
Licenses are as stated on the repository. 'other' is a creator's own license: check its commercial terms there. API price is the representative rate Artificial Analysis publishes, not a median across providers.
How this is calculatedExchange rate, how a plan is chosen, what is excluded, and the disclaimer
How this is calculated
- 1KRW is converted at 1 USD = 1,324.7 KRW as at 2026-10-25. The rate is refreshed daily.
- 2Monthly cost is the monthly-billing price of the cheapest plan that covers the usage entered. Overage is added only where the official page states an overage rate; a plan whose allowance is exceeded with no stated overage is left out of the ranking.
- 3Credit-based tools are converted using the credits-per-unit figure printed on the official page. Tools whose credit cost varies by model, with no official conversion published, are excluded from the ranking and listed separately.
- 4The per-unit price is the plan's monthly price divided by its monthly allowance. Unlimited plans show none, as there is nothing to divide by.
- 5A 원고지 sheet is 200 characters; an A4 page is taken as 1,800 characters at 10pt with 160% line spacing.
- 6Prices are as published by the vendor. VAT, card issuers' foreign-transaction fees and the rate applied on a KRW settlement are excluded. Regional pricing and temporary promotions may not be reflected.
- 7Paying in dollars by card means your issuer adds its own exchange rate and a foreign-transaction fee, typically 1-2% of the amount. That figure depends on the issuer and the moment of payment, so Wolmma cannot know it and does not include it.
Disclaimer
Wolmma is an information site that collects prices from each tool's official pricing page. We may hold an affiliate relationship with a vendor, but we are neither the seller nor its agent.
We verify as carefully as we can, but a displayed price, allowance or conversion may differ from what you are actually charged, because of plan changes, regional policy, promotions or tax. Confirm the final amount on the official page before you pay.
These figures are reference information, not a recommendation or advice to buy. Wolmma accepts no liability for loss arising from a purchase or contract entered into on the basis of this information.
What your card is actually billed will differ from the figure shown, according to your issuer's exchange rate, its foreign-transaction fee and local tax. None of that is within our sight or control, and we accept no liability for it. Payment, refunds and plan changes are a contract between you and the vendor, to which the Company is not a party.
If a price is wrong, write to connect@aiyogeum.com. We check it, correct it, and record the correction in the change log.