Skip to main content
BenchLM
Data
Recommendation
Ranking data as of

Open-Source LLM Leaderboard 2026

Kimi K3 is first on the November 2026 open-weight ranking at 68.4. Kimi K3.1 Pro scores 70.4 on a single-source estimate within 2 points, so it ranks second. Ling 3.1 Flash ranks third at 64.7. The license directory below separates OSI-approved licenses from community terms.

Bottom line: use the live table above for capability order, then check evidence status, license terms, memory requirements, and serving cost before choosing a deployment.

Citable stat

94 open-weight models are ranked as of November 13, 2026; Kimi K3 leads at 68.4/100.

Full Rankings (94 ranked)

1
Kimi K3
Moonshot AI·Open Weight·1.05M

68.4

BenchAlign v5.8

Supported

90% interval 63.99–72.86

vs #2
2
Kimi K3.1 Pro
Moonshot AI·Open Weight·1.05M

70.4

BenchAlign v5.8

Estimated

Conditional range 56.01–84.71

vs #3
3
Ling 3.1 Flash
InclusionAI·Open Weight·262K

64.7

BenchAlign v5.8

Estimated

Conditional range 50.31–79.00

vs #4
4
DeepSeek V4.1 Flash 1017
DeepSeek·Open Weight·1M

64.5

BenchAlign v5.8

Supported

90% interval 59.05–69.92

vs #5
5
Qwen3.8 Max
Alibaba·Open Weight·1M

63.6

BenchAlign v5.8

Supported

90% interval 54.85–72.29

vs #6
6
dots3-note Preview
Dots Studio·Open Weight·512K

62.8

BenchAlign v5.8

Estimated

Conditional range 48.42–77.12

vs #7
7
Ornith-1.5-397B
Ornith AI·Open Weight·262K

61.4

BenchAlign v5.8

Estimated

Conditional range 47.04–75.74

vs #8
8
GLM-5.3
Z.AI·Open Weight·1M

61

BenchAlign v5.8

Supported

90% interval 51.19–70.72

vs #9
9
Sage-A1.5
InternScience·Open Weight·262K

60.7

BenchAlign v5.8

Estimated

Conditional range 46.37–75.07

vs #10
10
Hy4 preview
Tencent·Open Weight·1M

60.4

BenchAlign v5.8

Estimated

Conditional range 46.04–74.74

vs #11
11
DeepSeek V4 Pro 0813
DeepSeek·Open Weight·1M

62.3

BenchAlign v5.8

Estimated

Conditional range 47.93–76.63

vs #12
12
GLM-5.2
Z.AI·Open Weight·1M

59.7

BenchAlign v5.8

Supported

90% interval 51.45–67.97

vs #13
13
MiMo-V2.6-Pro
Xiaomi·Open Weight·1M

57.3

BenchAlign v5.8

Estimated

Conditional range 47.53–66.97

vs #14
14
GLM-5.3-Flash
Z.AI·Open Weight·1M

56.8

BenchAlign v5.8

Estimated

Conditional range 47.11–66.56

vs #15
15
Kimi K2.6
Moonshot AI·Open Weight·256K

56.6

BenchAlign v5.8

Estimated

Conditional range 46.92–66.36

vs #16
16
GLM-5.1
Z.AI·Open Weight·203K

55.5

BenchAlign v5.8

Supported

90% interval 44.38–66.57

vs #17
17
Step 5 Preview
StepFun·Open Weight·1M

55.5

BenchAlign v5.8

Estimated

Conditional range 45.74–65.18

vs #18
18
Kimi K2.5 (Reasoning)
Moonshot AI·Open Weight·128K

56.6

BenchAlign v5.8

Estimated

Conditional range 42.28–70.98

vs #19
19
MiMo-V2.6-Flash
Xiaomi·Open Weight·1M

54.5

BenchAlign v5.8

Estimated

Conditional range 44.81–64.25

vs #20
20
GLM-5
Z.AI·Open Weight·200K

54.2

BenchAlign v5.8

Supported

90% interval 43.35–65.07

vs #21
21
Qwen3.8-27B
Alibaba·Open Weight·262K

54

BenchAlign v5.8

Supported

90% interval 41.45–66.63

vs #22
22
Qwen3.5 397B
Alibaba·Open Weight·128K

54.7

BenchAlign v5.8

Estimated

Conditional range 40.35–69.05

vs #23
23
Kimi K2.5
Moonshot AI·Open Weight·256K

52.6

BenchAlign v5.8

Supported

90% interval 43.64–61.47

vs #24
24
Mistral Large 4
Mistral·Open Weight·1M

52.1

BenchAlign v5.8

Supported

90% interval 41.83–62.35

vs #25
25
Qwen3.8-Flash-Next
Alibaba·Open Weight·262K

54

BenchAlign v5.8

Estimated

Conditional range 39.69–68.39

vs #26
26
Kimi K2.7 Code
Moonshot AI·Open Weight·256K

53.7

BenchAlign v5.8

Estimated

Conditional range 39.31–68.01

vs #27
27
Ternary Bonsai 2 27B
Prism ML·Open Weight·262K

51.6

BenchAlign v5.8

Estimated

Conditional range 37.29–65.99

vs #28
28
Qwen3.5-122B-A10B
Alibaba·Open Weight·262K

52.2

BenchAlign v5.8

Estimated

Conditional range 37.83–66.53

vs #29
29
Inkling-Small
Thinking Machines Lab·Open Weight·1M

49.9

BenchAlign v5.8

Supported

90% interval 33.53–66.24

vs #30
30
DeepSeek V3.2
DeepSeek·Open Weight·128K

49.8

BenchAlign v5.8

Supported

90% interval 36.21–63.41

vs #31
31
MiniMax M3
MiniMax·Open Weight·1M

49.4

BenchAlign v5.8

Supported

90% interval 34.30–64.51

vs #32
32
Nemotron 3.5 Super VL 120B A12B
NVIDIA·Open Weight·1M

48.8

BenchAlign v5.8

Estimated

Conditional range 34.42–63.12

vs #33
33
Nemotron 3 Ultra
NVIDIA·Open Weight·1M

48.7

BenchAlign v5.8

Estimated

Conditional range 34.38–63.08

vs #34
34
Muse Spark Open
Meta·Open Weight·1M

50.7

BenchAlign v5.8

Estimated

Conditional range 36.35–65.05

vs #35
35
Inkling
Thinking Machines Lab·Open Weight·1M

48.2

BenchAlign v5.8

Supported

90% interval 30.79–65.67

vs #36
36
Qwen3.6-27B
Alibaba·Open Weight·262K

48

BenchAlign v5.8

Estimated

Conditional range 33.68–62.38

vs #37
37
Qwen3.5-27B
Alibaba·Open Weight·262K

47.9

BenchAlign v5.8

Estimated

Conditional range 38.20–57.65

vs #38
38
GLM-4.7
Z.AI·Open Weight·200K

47.7

BenchAlign v5.8

Supported

90% interval 35.37–60.11

vs #39
39
MiniMax M2.7
MiniMax·Open Weight·200K

46.8

BenchAlign v5.8

Supported

90% interval 35.08–58.50

vs #40
40
Qwen3.5-35B-A3B
Alibaba·Open Weight·262K

45.2

BenchAlign v5.8

Supported

90% interval 34.04–56.40

vs #41
41
Beam 1.0 501B-A23B
Reflection AI·Open Weight·1M

43.2

BenchAlign v5.8

Estimated

Conditional range 33.51–52.95

vs #42
42
Ling 3.0 Flash
InclusionAI·Open Weight·262K

45.1

BenchAlign v5.8

Estimated

Conditional range 30.75–59.45

vs #43
43
Step 3.7 Flash
StepFun·Open Weight·256K

44.5

BenchAlign v5.8

Estimated

Conditional range 30.18–58.88

vs #44
44
Step 3.5 Flash
StepFun·Open Weight·256K

42.2

BenchAlign v5.8

Estimated

Conditional range 32.50–51.94

vs #45
45
GLM-4.6
Z.AI·Open Weight·200K

41.9

BenchAlign v5.8

Estimated

Conditional range 32.19–51.64

vs #46
46
Nemotron 3.5 Bolt 74B-A9B
NVIDIA·Open Weight·256K

41.6

BenchAlign v5.8

Estimated

Conditional range 27.25–55.95

vs #47
47
Gemma 4 26B A4B
Google·Open Weight·256K

41.2

BenchAlign v5.8

Estimated

Conditional range 31.51–50.95

vs #48
48
Trinity-Large-Thinking
Arcee AI·Open Weight·512K

43.2

BenchAlign v5.8

Estimated

Conditional range 28.87–57.57

vs #49
49
DeepSeek-R1
DeepSeek·Open Weight·128K

40.7

BenchAlign v5.8

Supported

90% interval 26.34–54.98

vs #50
50
DeepSeek V3.1
DeepSeek·Open Weight·128K

40.3

BenchAlign v5.8

Supported

90% interval 23.63–56.95

vs #51
51
Qwen3.6-35B-A3B
Alibaba·Open Weight·262K

42

BenchAlign v5.8

Estimated

Conditional range 27.65–56.35

vs #52
52
MiMo-V2-Flash
Xiaomi·Open Weight·256K

39.8

BenchAlign v5.8

Estimated

Conditional range 30.08–49.52

vs #53
53
Spark-X2.6-4B
SparkLLM·Open Weight·1M

39

BenchAlign v5.8

Estimated

Conditional range 24.67–53.37

vs #54
54
GLM-4.7-Flash
Z.AI·Open Weight·200K

38.7

BenchAlign v5.8

Estimated

Conditional range 28.99–48.44

vs #55
55
DeepSeek V3.1 (Reasoning)
DeepSeek·Open Weight·128K

38.7

BenchAlign v5.8

Estimated

Conditional range 28.94–48.38

vs #56
56
GPT-OSS 20B
OpenAI·Open Weight·128K

36.9

BenchAlign v5.8

Estimated

Conditional range 27.14–46.58

vs #57
57
DeepSeek V3 0324
DeepSeek·Open Weight·128K

35.9

BenchAlign v5.8

Supported

90% interval 19.45–52.36

vs #58
58
LFM2.5-24B-A2B
LiquidAI·Open Weight·131K

34.4

BenchAlign v5.8

Estimated

Conditional range 20.08–48.78

vs #59
59
Gemma 4 12B
Google·Open Weight·256K

34.1

BenchAlign v5.8

Estimated

Conditional range 19.75–48.45

vs #60
60
Mistral Large 3
Mistral·Open Weight·128K

33.3

BenchAlign v5.8

Estimated

Conditional range 23.61–43.05

vs #61
61
Ornith-1.5-35B-A3B
Ornith AI·Open Weight·262K

33

BenchAlign v5.8

Estimated

Conditional range 18.69–47.39

vs #62
62
Ornith-1.5-9B
Ornith AI·Open Weight·262K

32.5

BenchAlign v5.8

Estimated

Conditional range 18.13–46.83

vs #63
63
ZAYA1-8B
Zyphra·Open Weight·131K

32

BenchAlign v5.8

Estimated

Conditional range 17.68–46.38

vs #64
64
GPT-OSS 120B
OpenAI·Open Weight·128K

31.9

BenchAlign v5.8

Supported

90% interval 13.51–50.31

vs #65
65
DeepSeek V3
DeepSeek·Open Weight·128K

31.5

BenchAlign v5.8

Supported

90% interval 15.15–47.92

vs #66
66
Nemotron 3 Nano Omni 30B A3B
NVIDIA·Open Weight·256K

31.4

BenchAlign v5.8

Estimated

Conditional range 17.05–45.75

vs #67
67
Ling 2.6 Flash
InclusionAI·Open Weight·262K

33.1

BenchAlign v5.8

Estimated

Conditional range 18.73–47.43

vs #68
68
DeepSeek R1 Distill Qwen 32B
DeepSeek·Open Weight·128K

30.9

BenchAlign v5.8

Estimated

Conditional range 16.50–45.20

vs #69
69
Muse Glimmer 30B
Meta·Open Weight·131K

32.2

BenchAlign v5.8

Estimated

Conditional range 17.83–46.53

vs #70
70
Sarvam 105B
Sarvam·Open Weight·128K

29.4

BenchAlign v5.8

Estimated

Conditional range 15.02–43.72

vs #71
71
Gemma 4 31B
Google·Open Weight·256K

28.8

BenchAlign v5.8

Estimated

Conditional range 19.09–38.53

vs #72
72
Qwen2.5-72B
Alibaba·Open Weight·128K

28.5

BenchAlign v5.8

Estimated

Conditional range 18.77–38.22

vs #73
73
Gemma 3 27B
Google·Open Weight·32K

28.4

BenchAlign v5.8

Supported

90% interval 9.66–47.17

vs #74
74
Llama 3.1 405B
Meta·Open Weight·128K

28

BenchAlign v5.8

Estimated

Conditional range 18.24–37.68

vs #75
75
Llama 4 Maverick
Meta·Open Weight·1M

27.7

BenchAlign v5.8

Estimated

Conditional range 17.94–37.38

vs #76
76
Llama 4 Scout
Meta·Open Weight·10M

26.7

BenchAlign v5.8

Estimated

Conditional range 17.01–36.45

vs #77
77
Command A+
Cohere·Open Weight·128K

28.6

BenchAlign v5.8

Estimated

Conditional range 14.28–42.97

vs #78
78
Qwen2.5 Coder 32B Instruct
Alibaba·Open Weight·128K

26.3

BenchAlign v5.8

Estimated

Conditional range 16.55–35.99

vs #79
79
Mistral Medium 3.5 128B
Mistral·Open Weight·256K

28.2

BenchAlign v5.8

Estimated

Conditional range 13.84–42.54

vs #80
80
Granite 4.2 8B
IBM·Open Weight·128K

25.9

BenchAlign v5.8

Estimated

Conditional range 16.18–35.63

vs #81
81
Mistral Small 4
Mistral·Open Weight·256K

25.9

BenchAlign v5.8

Estimated

Conditional range 16.18–35.62

vs #82
82
Gemma 4 E4B
Google·Open Weight·128K

24.6

BenchAlign v5.8

Estimated

Conditional range 10.24–38.94

vs #83
83
LFM2.5-8B-A1B
LiquidAI·Open Weight·128K

23.9

BenchAlign v5.8

Estimated

Conditional range 9.51–38.21

vs #84
84
Gemma 4 E2B
Google·Open Weight·128K

23.8

BenchAlign v5.8

Estimated

Conditional range 9.40–38.10

vs #85
85

23.1

BenchAlign v5.8

Estimated

Conditional range 8.78–37.48

vs #86
86
LFM2.5-2.6B
LiquidAI·Open Weight·128K

23.1

BenchAlign v5.8

Estimated

Conditional range 8.74–37.44

vs #87
87
Phi-4
Microsoft·Open Weight·16K

22.9

BenchAlign v5.8

Estimated

Conditional range 13.15–32.59

vs #88
88
Sarvam 30B
Sarvam·Open Weight·64K

22

BenchAlign v5.8

Estimated

Conditional range 7.60–36.29

vs #89
89
Llama 3 70B
Meta·Open Weight·128K

20.7

BenchAlign v5.8

Estimated

Conditional range 10.98–30.43

vs #90
90
LFM3-1.7B
LiquidAI·Open Weight·256K

19.1

BenchAlign v5.8

Estimated

Conditional range 4.77–33.46

vs #91
91
Mixtral 8x22B Instruct v0.1
Mistral·Open Weight·64K

18.4

BenchAlign v5.8

Estimated

Conditional range 8.72–28.16

vs #92
92
Nemotron-4 15B
NVIDIA·Open Weight·32K

9.9

BenchAlign v5.8

Estimated

Conditional range 0.19–19.63

vs #93
93
Mistral 7B v0.3
Mistral·Open Weight·32K

7.2

BenchAlign v5.8

Estimated

Conditional range 0.00–16.90

vs #94
94
MiniCPM5-1B
OpenBMB·Open Weight·131K

4.2

BenchAlign v5.8

Estimated

Conditional range 0.00–18.59

Current position

Kimi K3 leads the live open-weight ranking at 68.4 with Supported evidence.

Kimi K3.1 Pro ranks #2 at 70.4 with Estimated evidence.

Ling 3.1 Flash ranks #3 at 64.7 with Estimated evidence.

How to choose

Key Takeaways

The top model is Kimi K3 by Moonshot AI with a BenchAlign v5.8 score of 68.4 and Supported evidence.

The best open-weight model is Kimi K3 at position #1.

94 models are included in this ranking.

Score in Context

What these scores mean

Open-weight models are ranked by the same public BenchAlign v5.8 overall score as proprietary models. Supported and Estimated describe the evidence behind each position; they are not separate leaderboards.

Known limitations

Open weight is not the same as OSI-approved open source. The ranking does not score license restrictions, memory requirements, serving throughput, fine-tuning support, or the engineering cost of operating the model.

About this ranking

Ranking data as of November 13, 2026

This page is the canonical open-weight ranking. It uses the same public BenchAlign v5.8 overall lane as the main leaderboard, then filters to downloadable model weights. The score compares measured capability; it does not decide whether a license is permissive, a model fits your hardware, or self-hosting beats an API on cost. Use the linked decision guide for those deployment questions.

This is the public BenchAlign v5.8 overall lane filtered to open-weight models. The table orders point estimates. Evidence labels describe source support; conditional ranges do not establish rank confidence.

The open-weight slice starts with Kimi K3, followed by Kimi K3.1 Pro and Ling 3.1 Flash. All rows use the public BenchAlign v5.8 projection contract. Evidence badges and score intervals matter as much as small point gaps because public source coverage is uneven.

Every ranked model publishes downloadable weights, but that does not make every license OSI-approved or every deployment practical. Use the directory below to check license terms, quantization, and reference hardware before shortlisting a model.

This ranking uses the public BenchAlign v5.8 overall contract filtered to open-weight models. For detailed model profiles, click any model name above. To compare two specific models head-to-head, use the "vs #" links.

Deployment evidence

Compare license, size, and deployment

The performance table ranks every eligible open-weight row. This smaller directory includes only models with a deployment record in the self-host catalog, so license and hardware claims remain auditable instead of being filled from family names.

Open weight is an access category, not a license verdict. “OSI-approved” appears only when the deployment catalog records MIT or Apache 2.0. Community and custom licenses stay separate, even when their weights are downloadable.

Showing 67 of 67 deployment-documented rows

Deployment catalog checked 2026-11-04.

This table is the sourced deployment subset, not the complete performance ranking. A missing row means the deployment catalog is incomplete, not that the model cannot be self-hosted.

Questions

What is the best open source LLM right now?

Kimi K3 is first at 68.4, as of November 13, 2026. Kimi K3.1 Pro scores 70.4 on a single-source estimate within 2 points, so it ranks second. Ling 3.1 Flash is third at 64.7.

Which LLMs are open source?

As of November 13, 2026, 94 open-weight models are ranked here. The five highest are Kimi K3 (Moonshot AI), Kimi K3.1 Pro (Moonshot AI), Ling 3.1 Flash (InclusionAI), DeepSeek V4.1 Flash 1017 (DeepSeek) and Qwen3.8 Max (Alibaba). Each publishes weights you can download. That isn’t the same as an OSI-approved open-source license, so check the license directory on this page before building on one.

Are open source LLMs as good as GPT or Claude?

Not at the top of the overall ranking, as of November 13, 2026. The highest-ranked open-weight model, Kimi K3, scores 68.4 and sits at #17 overall, where the first row scores 93.7. On single tasks an open model can match a closed one, and control over deployment, privacy and serving cost can still make it the better choice.

What is the best open source LLM for coding?

Use the open-weight rows on the live coding leaderboard. Coding order differs from overall order, and a single HumanEval or LiveCodeBench result should not be treated as a complete coding verdict.

Can I run these models locally?

The smaller open-weight rows run on a single consumer GPU with 4-bit quantization; frontier-size models need multi-GPU rigs or high-memory Apple Silicon. The local LLM guide breaks the rankings down by VRAM tier, and the Ollama guide includes pull commands and size estimates per model.

Which labs make the best open-weight models?

The ten highest-ranked open-weight models as of November 13, 2026 come from nine labs: Moonshot AI (2), InclusionAI (1), DeepSeek (1), Alibaba (1), Dots Studio (1), Ornith AI (1), Z.AI (1), InternScience (1) and Tencent (1). Strengths differ by task, so check the category leaderboards before choosing by lab.

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 7,000+ subscribers.

One email each week. Unsubscribe anytime.