Best Open-Source / Open-Weight AI
| # | Model | Region | Tier | Price in / out | Context | Strengths | |
|---|---|---|---|---|---|---|---|
| 1 | Kimi K3Moonshot AI · open | 🇨🇳 CN | Frontier | $3 / $15 | 1M | China's top-priced reasoning model; open-weight | src |
| 2 | DeepSeek V4 ProDeepSeek · open | 🇨🇳 CN | Pro | $1.32 / $3.96 | 1M | Low price; 50% off-peak Peak cache-miss rate $1.32/$3.96 rechecked 2026-11-06 on DeepSeek API docs (unchanged). Open-weights status: verify. | src |
| 3 | Muse Spark OpenMeta · open | 🇺🇸 US | Pro | $1.25 / $4.25 | 1M | Meta's open-weight coding & agent model Same $1.25/$4.25 at all 5 OpenRouter hosts. Weights under Meta's Muse Model License (limited commercial terms). | src |
| 4 | GLM-5.3Zhipu / Z.ai · open | 🇨🇳 CN | Pro | $1.4 / $4.4 | 1M | Strong value open-weight model | src |
| 5 | DeepSeek V4.1 FlashDeepSeek · open | 🇨🇳 CN | Strong | $0.3 / $1.2 | 1M | Very cheap; half off-peak Open-weights status: verify | src |
| 6 | gpt-oss-120BOpenAI (via Together) · open | 🇺🇸 US | Strong | $0.15 / $0.6 | 131K | Open-weights OpenAI model | src |
| 7 | Nemotron 3.5 Super VLNVIDIA · open | 🇺🇸 US | Strong | $0.15 / $0.5 | 1M | Open multimodal MoE (12B active) Lowest OpenRouter host price (Crusoe); other hosts $0.15-$0.25 in. | src |
| 8 | MiniMax-M3MiniMax · open | 🇨🇳 CN | Strong | $0.3 / $1.2 | 1M | Very cheap (permanent 50% off; 2x above 512K input) Open-weights status: verify | src |
| 9 | MiniMax-M3.1-FlashMiniMax · open | 🇨🇳 CN | Strong | $0.15 / $0.6 | 1M | Cheap multimodal flash for agents, coding and computer use; 1M context Official MiniMax API list price $0.15/$0.60 rechecked 2026-11-06 (unchanged from 2026-11-05). Free launch on OpenRouter ended 2026-10-22. Open-weights status: verify. | src |
| 10 | Mistral Small 4Mistral · open | 🇪🇺 EU | Lite | $0.15 / $0.6 | 262K | Cheap, efficient Open-weights status: verify | src |
| 11 | GLM-5.3-FlashZhipu / Z.ai · open | 🇨🇳 CN | Lite | $0.15 / $0.5 | 1M | Cheapest GLM tier Open-weights status: verify | src |
What is the best AI model for open-weight use in 2026?
Of the 11 models we list for open-weight use, the highest editorial tier (Frontier) is held by Kimi K3 (Moonshot AI, $3 input / $15 output per 1M tokens). Prices as of 2026-11-19 from the linked provider pages.
What is the cheapest model in this list?
Nemotron 3.5 Super VL by NVIDIA has the lowest blended price here: $0.15 input / $0.5 output per 1M tokens. Blended price weights input 3:1 over output.
How many models are on this page and where are they from?
11 models: 3 from the US, 7 from China and 1 from Europe.
Which of these models are open-weight?
Kimi K3, DeepSeek V4 Pro, Muse Spark Open, GLM-5.3, DeepSeek V4.1 Flash, gpt-oss-120B, Nemotron 3.5 Super VL, MiniMax-M3, MiniMax-M3.1-Flash, Mistral Small 4, GLM-5.3-Flash (11 of 11) publish open weights.
How is this list ranked?
By AIRaceBoard's editorial capability tier, then by price. These are editorial picks, not benchmark scores; every price links to its source.
Model pricing pages
- Kimi K3 API pricing
- DeepSeek V4 Pro API pricing
- Muse Spark Open API pricing
- GLM-5.3 API pricing
- DeepSeek V4.1 Flash API pricing
- gpt-oss-120B API pricing
- Nemotron 3.5 Super VL API pricing
- MiniMax-M3 API pricing
- MiniMax-M3.1-Flash API pricing
- Mistral Small 4 API pricing
- GLM-5.3-Flash API pricing
Best Chinese AI models (mostly open-weight) · Compare models head to head · Cost calculator