Home  /  Blog  /  Best open-weight AI models
League analysis

The best open-weight AI models — November 2026, ranked by League Score.

The best open-weight AI models are the ones you can download, self-host and fine-tune — and in our database 31 of the 64 models we track qualify. This post ranks the scored ones by League Score, measures the gap to the closed frontier, prices the leader against the overall #1 on a real workload, and flags where a score rests on thin benchmark coverage.

By AI Pro League · Published

Best open-weight AI models ranked by League Score in the AI Pro League database, November 2026

Best open-weight AI models, by our AI model rankings: Inkling 2 Small leads the open field at 87.0, and it is not close to the closed frontier — the overall leader, Claude Fable 5.2, sits at 100.0, a gap of 13.0 points. Of the 31 open-weights models we track, 23 carry a League Score (the rest have no recorded benchmark components yet, so they are unranked rather than scored zero). The full open table lives in the Open-Weight Division, and our picks page ranks the same field in best open-source AI models.

The best open-weight AI models, ranked

Top 10 open-weight models by League Score, computed from the AI Pro League database
#ModelMakerLeague ScoreCoverageBlended $/1MLicence
1Inkling 2 SmallThinking Machines Lab87.015%—Apache 2.0
2DeepSeek V4.1 FlashDeepSeek84.455%$0.52MIT
3Kimi K3.1 ProMoonshot AI75.455%$4.00Kimi K3.1 Pro License
4Kimi K3Moonshot AI74.670%$6.00Kimi K3 License
5DeepSeek V4.1 Flash 1017DeepSeek73.755%$0.52MIT
6Inkling SmallThinking Machines Lab71.715%—Apache 2.0
7Kimi K2.6Moonshot AI69.735%$1.71Modified MIT
8Qwen3.8-MaxAlibaba Qwen68.565%$3.00Qwen3.8-Max License
9DeepSeek V4 Pro (0813)DeepSeek63.260%$1.98MIT
10Qwen3.5-397B-A17BAlibaba Qwen60.735%—Apache-2.0

The shape of the field matters as much as the order. The median scored open-weight model sits at 53.1, against an overall median of 69.5 across all 51 scored models — the open field is top-heavy: a strong leading group, then a long tail. The leader, Inkling 2 Small, does it on 15% benchmark coverage and carries no API price in our database — its weights are downloadable under Apache 2.0, so its cost is hardware, not tokens.

How far behind the closed frontier is the best open-weight model?

13.0 League points, on the numbers above — Inkling 2 Small at 87.0 versus Claude Fable 5.2 at 100.0. Read that gap with the coverage column in hand: the leader’s score covers 15% of our five-benchmark set, the overall leader’s covers 40%. Neither is a full record, and a wider benchmark record could move either score — the honest statement is that on the benchmarks both models have actually been measured on, the closed frontier is ahead by a clear but not overwhelming margin.

What the best open-weight AI models cost — the worked example

The open leader itself carries no recorded API price in our database, so price the highest-ranked open model that does — DeepSeek V4.1 Flash (open #2) — against the overall leader on one workload: 10M input tokens and 2M output tokens, at the rates in our AI model pricing index (DeepSeek V4.1 Flash: $0.30 / $1.20 per 1M, checked Oct 18, 2026 against DeepSeek’s official pricing page; Claude Fable 5.2: $8.00 / $40.00 per 1M):

Workload maths: 10M input + 2M output, top priced open model vs overall leader
ModelInput costOutput costTotal
DeepSeek V4.1 Flash (open #2)$3.00$2.40$5.40
Claude Fable 5.2 (overall #1)$80.00$80.00$160.00

The overall leader costs 30× as much on this workload — and for the open model, the hosted API is only one option: the weights can be self-hosted, where the bill becomes hardware instead of tokens. That option, not the score gap, is why the open field keeps its audience. Our prompt caching pricing explainer covers the cached rates that change this maths on repeated-context work.

Coverage: why some open-weight scores deserve a second look

The deepest benchmark record in the open field belongs to Nemotron 3 Ultra 550B-A55B: 85% coverage — yet it ranks only #16 among open models at 37.8, because its recorded components are genuinely mid-field. Depth of record and height of score are different things, and this table keeps both visible: a high score on a thin record (several top-10 entries sit at 35% coverage or below) is a claim about the benchmarks that exist, not the ones that don’t. Our benchmark explainers — LiveBench, SWE-bench Verified, GPQA Diamond — set out what each component does and does not prove.

Which licence do the best open-weight AI models actually ship under?

“Open-weight” is not one licence. The 23 scored open models in our database ship under 12 different licences — Apache 2.0, Apache-2.0, Kimi K3 License, Kimi K3.1 Pro License and more — and the terms decide what self-hosting actually permits: some are permissive (MIT, Apache-2.0), others are community or ecosystem licences with conditions attached. The licence column in the table above is part of the ranking, not fine print: check it before a model’s score talks you into building on weights you may not be able to ship.