Best open-weight LLM for coding

Qwen3.8 Flash leads: 87.4
models ranked: 60
downloads of #1, 30 days: 2.49M
lead over #2: +4.1

Coding index · read October 20

New DeepSeek V4.1 Flash 1017 October 17
New E Glimpse October 17
New Mercury 3 October 16
New R Beam 1.0 501B A23B October 16

As of October 20, 2026, Qwen3.8 Flash leads 60 open-weight models with a coding index of 87.4, ahead of DeepSeek V4.1 Flash 1017 (83.3) and DeepSeek V4.1 Flash (77.2).

This list narrows the best LLM for coding ranking.

Top 20 of 60, higher is better

Embed this data

Ordered by index, highest first. The dots show how many benchmarks stand behind each score.

Best open-weight LLM for coding
Rank	Model	Coding index	Weights	Output $/Mtok
1	Qwen3.8 Flash Qwen	87.4	Qwen/Qwen3.8-Flash-Next	$0.47
2	DeepSeek V4.1 Flash 1017 DeepSeek	83.3	deepseek-ai/DeepSeek-V4.1-Flash-1017	$0.60
3	DeepSeek V4.1 Flash DeepSeek	77.2	deepseek-ai/DeepSeek-V4.1-Flash	$0.0112
4	Qwen3.8 27B Qwen	73.6	Qwen/Qwen3.8-27B	$2.20
5	GLM 5.3 Z.ai	73.1	zai-org/GLM-5.3	$1.40
6	MiMo-V2.6-Pro Xiaomi	72.6	XiaomiMiMo/MiMo-V2.6-Pro-RL	$0.87
7	Step 5 Preview StepFun	71.4	stepfun-ai/Step-5-Preview	$2.70
8	R1 0528 DeepSeek	71.4	deepseek-ai/DeepSeek-R1-0528	$2.15
9	GLM 5.3 Flash Z.ai	71.0	zai-org/GLM-5.3-Flash	$0.16
10	DeepSeek V3 DeepSeek	70.2	deepseek-ai/DeepSeek-V3	$1.0287
11	Nemotron 3.5 Super VL NVIDIA	65.5	nvidia/NVIDIA-Nemotron-3.5-Super-VL	$0.50
12	Kimi K2 0711 Moonshot AI	64.7	moonshotai/Kimi-K2-Instruct	$2.30
13	GLM 4.5 Z.ai	64.2	zai-org/GLM-4.5	$2.20
14	Kimi K2 Thinking Moonshot AI	63.4	moonshotai/Kimi-K2-Thinking	$2.50
15	Muse Spark Open Meta	62.1	meta-models/Muse-Spark-Open-816B-A42B	$4.25
16	MiMo-V2.6-Flash Xiaomi	61.9	XiaomiMiMo/MiMo-V2.6-Flash-RL	$0.28
17	Kimi K3 Moonshot AI	61.8	moonshotai/Kimi-K3	$11.936
18	Hy4 preview Tencent	61.6	tencent/Hy4-preview	$2.501
19	Qwen3 Coder 30B A3B Instruct Qwen	60.4	Qwen/Qwen3-Coder-30B-A3B-Instruct	$0.28
20	GLM 5 Z.ai	59.7	zai-org/GLM-5	$1.92

How this list is built

Generated from the dataset on every refresh. Nobody picks the order by hand.

Scored models that OpenRouter links to a public Hugging Face repository, which is where their weights are published, ordered as the ranking orders them. We do not check the license; read it on the repository page.

OpenRouter, fetched October 20, 2026
Aider polyglot leaderboard, fetched October 20, 2026
LMArena WebDev, fetched October 20, 2026
SWE-bench (swebench.com), fetched October 20, 2026
LiveBench, fetched October 20, 2026, board updated October 14, 2026

Other lists

Rankings beyond coding · Methodology