← the full leaderboard · open weights
The best open-weights AI models, ranked.
Kimi K3.1 Pro is the highest-ranked open-weights model on the ArtificialWatch Bench Index (#13 of 50). Open-weights models ranked against every model on the index, with license, size and Hugging Face repo. Updated daily.
Scores as published by each board; table composed November 11, 2026.
Get the next launch alert, free →
Open-weights leaderboard
| # | Model | AW index | Coding index | Size | License | Weights | API price / 1M |
|---|---|---|---|---|---|---|---|
| 1 | Kimi K3.1 Pro · Moonshot | 67.9 · #13 of 50 | 31 · #35 | 2.8T | custom (kimi-k3.1-pro) | Hugging Face | $2 / $10 |
| 2 | DeepSeek V4.1 Flash · DeepSeek | 61.2 · #16 of 50 | 71.4 · #11 | 763B | MIT | Hugging Face | — |
| 3 | Hy4 · Tencent | 59.3 · #18 of 50 | — | 780B | Apache 2.0 | Hugging Face | — |
| 4 | GLM-5.3 · Zhipu AI | 48.1 · #23 of 50 | 51.5 · #20 | 753B | custom (glm-5.3) | Hugging Face | $1.40 / $4.40 |
| 5 | MiMo-V2.6 Pro · Xiaomi | 42 · #28 of 50 | 67.3 · #14 | 1.02T | MIT | Hugging Face | $0.435 / $0.87 |
| 6 | Kimi K3 · Moonshot | 37 · #30 of 50 | 65.3 · #16 | 2.8T | custom (kimi-k3) | Hugging Face | $3 / $15 |
| 7 | MiniMax M3.1 Flash · MiniMax | 35.2 · #31 of 50 | 30.1 · #38 | 220B | custom (minimax-community) | Hugging Face | $0.15 / $0.60 |
| 8 | Step 5 Preview · StepFun | 31.7 · #32 of 50 | 33.3 · #33 | 604B | custom (stepfun-model-license) | Hugging Face | — |
| 9 | DeepSeek V4 Pro · DeepSeek | 23.3 · #39 of 50 | 39.5 · #27 | 1.65T | MIT | Hugging Face | $0.66 / $1.98 |
| 10 | GLM-5.3 Flash · Zhipu AI | 22.2 · #40 of 50 | 40.7 · #26 | 321B | MIT | Hugging Face | $0.15 / $0.50 |
| 10 | Mistral Large 4 · Mistral | 22.2 · #40 of 50 | 14.3 · #45 | 1.05T | custom (mistral-ecosystem-open-weight) | Hugging Face | $0.70 / $2.10 |
| 12 | Muse Spark Open · Meta | 18.5 · #43 of 50 | 21.9 · #42 | 822B | custom (muse-model-license) | Hugging Face | — |
| 13 | GLM-5.2 · Zhipu AI | 14.3 · #44 of 50 | 30.6 · #36 | 753B | MIT | Hugging Face | $1.40 / $4.40 |
| 14 | Beam · Reflection AI | 6.9 · #47 of 50 | 24.5 · #40 | 501B | Apache 2.0 | Hugging Face | — |
| 15 | MiniMax M3 · MiniMax | 5 · #48 of 50 | 16.8 · #44 | 427B | custom (minimax-community) | Hugging Face | $0.30 / $1.20 |
| 16 | Inkling · Thinking Machines | 3.2 · #49 of 50 | 3.7 · #46 | 952B | Apache 2.0 | Hugging Face | — |
| 17 | Inkling Small · Thinking Machines | 1.9 · #50 of 50 | 3.6 · #47 | 266B | Apache 2.0 | Hugging Face | — |
AW index: the median of each model's percentile across the public leaderboards that score it, ranked against all 50 models on the full leaderboard. Coding index: the same rule over three coding boards (the coding leaderboard). Size: total parameters as Hugging Face counts the repo's weights. License: as the repo lists it. Prices are input / output per 1M tokens on the maker's own API.
The list as JSON or CSV (repo, licence, size, gate and ranks), free to reuse under CC BY 4.0 with a link to this page. More files on open data.
Best open-source LLM for coding
- DeepSeek V4.1 Flash · coding index 71.4 · #11 of 47 on the coding leaderboard
- MiMo-V2.6 Pro · coding index 67.3 · #14 of 47 on the coding leaderboard
- Kimi K3 · coding index 65.3 · #16 of 47 on the coding leaderboard
- GLM-5.3 · coding index 51.5 · #20 of 47 on the coding leaderboard
- GLM-5.3 Flash · coding index 40.7 · #26 of 47 on the coding leaderboard
- DeepSeek V4 Pro · coding index 39.5 · #27 of 47 on the coding leaderboard
- Step 5 Preview · coding index 33.3 · #33 of 47 on the coding leaderboard
- Kimi K3.1 Pro · coding index 31 · #35 of 47 on the coding leaderboard
- GLM-5.2 · coding index 30.6 · #36 of 47 on the coding leaderboard
- MiniMax M3.1 Flash · coding index 30.1 · #38 of 47 on the coding leaderboard
- Beam · coding index 24.5 · #40 of 47 on the coding leaderboard
- Muse Spark Open · coding index 21.9 · #42 of 47 on the coding leaderboard
- MiniMax M3 · coding index 16.8 · #44 of 47 on the coding leaderboard
- Mistral Large 4 · coding index 14.3 · #45 of 47 on the coding leaderboard
- Inkling · coding index 3.7 · #46 of 47 on the coding leaderboard
- Inkling Small · coding index 3.6 · #47 of 47 on the coding leaderboard
Ordered by coding rank. The full AI coding leaderboard →
Quick answers
What is the best open-weights (open-source) AI model right now?
Kimi K3.1 Pro from Moonshot: the highest-ranked model with downloadable weights on the ArtificialWatch Bench Index, at 67.9 (#13 of 50 models, open and closed). Next are DeepSeek V4.1 Flash (61.2) and Hy4 (59.3). Its weights are on Hugging Face as moonshotai/Kimi-K3.1-Pro, license custom (kimi-k3.1-pro).
What is the best open-weights model for coding?
DeepSeek V4.1 Flash, at 71.4 on the ArtificialWatch coding index (#11 of 47 models, open and closed).
How far behind closed models are open-weights models?
The best open-weights model, Kimi K3.1 Pro, ranks #13 of 50 on the Bench Index at 67.9; 12 models rank above it, led by Claude Fable 5.2 at 100. The index is the median percentile across public leaderboards (0-100), so the gap is a gap in rank on those boards, not a measure of how much more a model can do.
Is open-weights the same as open source?
Not quite. Open-weights means the trained weights can be downloaded and run on your own hardware. The Open Source Initiative's definition also covers the license terms and detailed information about the training data. Several models here ship under their maker's own license rather than a standard one like MIT or Apache 2.0, and each row names the license its Hugging Face repo lists.
Which models count as open-weights here?
Models whose weights are public on Hugging Face and that at least 2 public leaderboards score: Kimi K3.1 Pro, DeepSeek V4.1 Flash, Hy4, GLM-5.3, MiMo-V2.6 Pro, Kimi K3, MiniMax M3.1 Flash, Step 5 Preview, DeepSeek V4 Pro, GLM-5.3 Flash, Mistral Large 4, Muse Spark Open, GLM-5.2, Beam, MiniMax M3, Inkling and Inkling Small. Each repo was last checked between November 6, 2026 and November 8, 2026.
More rankings
The full LLM leaderboard · The AI coding leaderboard · Every model's API price · Head-to-head comparisons
Use this chart
Writing about open-weights models? Paste this where you want the chart. It redraws when the table updates, and the credit line links back to this page.
<figure><a href="https://artificialwatch.com/open-weights"><img src="https://artificialwatch.com/api/aw/og/leaderboard/open" alt="The top five open-weights AI models on the ArtificialWatch bench index" width="600" height="315"></a><figcaption>Chart: <a href="https://artificialwatch.com/open-weights">ArtificialWatch</a>, CC BY 4.0</figcaption></figure>Reports from this data
Nov 6, 2026
8 of 17 open-weights AI models ship under a license of their own, not MIT or Apache, including 6 of the top 10Nov 4, 2026
Every open-weights date an AI lab has set since July was met: five promises, five kept, after a median wait of 12 daysNov 3, 2026
The best open-weight AI model ranks 12th, 30 points behind Claude Opus 5.5. The top six open models are all Chinese
Every report, from every dataset: data reports.
Hear when the next one drops. Free.
Free: a Chrome push and an email about 15 minutes after a new model answers on a public API. Members get a text or a phone call the moment a launch is confirmed.
- Free forever
- No card
- Unsubscribe in one click