Coding and software agents

Open coding model releases

A focused comparison of open-weight or open-source models positioned for coding, repository work, tool use, and software-engineering agents.

Open-weight coding models have closed much of the gap with proprietary frontier systems. This table compares the downloadable ones built for software work, ranked by their SWE-bench Verified claim — resolving real GitHub issues — then by recency.

Size and context matter as much as the headline score: a smaller model with a long context can be the better fit for whole-repository work on local hardware.

SWE-bench values are included only when the catalog has a sourced claim or verified entry. Vendor-reported scores remain claims until independently verified.

Model	Lab	License	Params	Context	SWE-bench	Released	Source
Nex-N2-Pro	Nex AGI	Apache-2.0	397B · 17B active	262K	80.8%	Jun 2, 2026	source
Kimi K2.6	Moonshot	Modified MIT	1T · 32B active	256K	80.2%	Mar 30, 2026	source
Hunyuan Hy3	Hunyuan	Apache-2.0	295B · 21B active	256K	78%	Jul 6, 2026	source
GLM-5	Z.ai	MIT	744B · 40B active	—	77.8%	Feb 11, 2026	source
Mistral Medium 3.5	Mistral	Mistral Research / Commercial	128B	256K	77.6%	Mar 18, 2026	source
Qwen3.6-27B	Qwen	Apache-2.0	27B	256K	77.2%	May 12, 2026	source
Kimi K3	Moonshot	Modified MIT	2.8T · 50B active	1.0M	76.8%	Jul 16, 2026	source
Kimi K2.5	Moonshot	Modified MIT	1T · 32B active	256K	76.8%	Jan 27, 2026	source
GLM-4.7	Z.ai	MIT	358B · 32B active	—	73.8%	Jan 8, 2026	source
Kimi K2 Thinking	Moonshot	Modified MIT	1T · 32B active	256K	71.3%	Nov 6, 2025	source
Qwen3-Coder-Next	Qwen	Apache-2.0	80B · 3B active	262K	70.6%	Feb 3, 2026	source
DeepSeek-V3.2	DeepSeek	MIT	685B · 37B active	128K	70%	Dec 1, 2025	source
Kimi K2 Instruct 0905	Moonshot	Modified MIT	1T · 32B active	256K	69.2%	Sep 5, 2025	source
DeepSeek-V3.1-Terminus	DeepSeek	MIT	685B · 37B active	128K	68.4%	Sep 22, 2025	source
Kimi K2 Instruct	Moonshot	Modified MIT	1T · 32B active	128K	65.8%	Jul 11, 2025	source
Kimi-Dev-72B	Moonshot	MIT	73B	—	60.4%	Jun 17, 2025	source
DeepSeek-R1-0528	DeepSeek	MIT	671B · 37B active	128K	57.6%	May 28, 2025	source
Seed-OSS-36B-Instruct	Seed	Apache-2.0	36B	512K	56%	Aug 20, 2025	source
MiniMax-M1-80k	MiniMax	Apache-2.0	456B · 45.9B active	1M	56%	Jun 16, 2025	source
Sarvam-105B	Sarvam	Apache-2.0	105B · 10.3B active	128K	45%	Mar 6, 2026	source
DeepSeek-V4-Flash-0731	DeepSeek	MIT	284B · 13B active	1M	-	Jul 31, 2026	source
Ling-3.0-flash	inclusionAI	Apache 2.0 (announced; weights not yet posted as of 2026-07-24)	124B · 5.1B active	262K	-	Jul 23, 2026	source
Laguna S 2.1	Poolside	OpenMDW-1.1	118B · 8B active	1M	-	Jul 21, 2026	source
Inkling	Thinking Machines	Apache-2.0	975B · 41B active	1.0M	-	Jul 15, 2026	source
Nemotron-Labs-3-Puzzle-75B-A9B	NVIDIA	OpenMDW-1.1	75.3B · 9.3B active	1M	-	Jul 6, 2026	source
Mistral frontier open-weight MoE (unnamed)	Mistral	Open weights	Undisclosed	—	-	Jul 6, 2026	source
Laguna XS 2.1	Poolside	OpenMDW-1.1	33B · 3B active	256K	-	Jul 2, 2026	source
LongCat-2.0	Meituan	MIT	1.6T · 48B active	1M	-	Jun 30, 2026	source
Kimi K2.7 Code	Moonshot	Modified MIT	1T · 32B active	262K	-	Jun 18, 2026	source
GLM-5.2	Z.ai	MIT	753B	1M	-	Jun 17, 2026	source
MiniMax-M3	MiniMax	MiniMax Community License	428B · 23B active	1M	-	Jun 16, 2026	source
DiffusionGemma 26B-A4B	DeepMind	Apache-2.0	25.2B · 3.8B active	256K	-	Jun 10, 2026	source
North Mini Code 1.0	Cohere	Apache-2.0	30B · 3B active	256K	-	Jun 9, 2026	source
Nemotron 3 Ultra 550B-A55B	NVIDIA	Nemotron Open Model License	550B · 55B active	1M	-	Jun 4, 2026	source
Gemma 4 12B	DeepMind	Apache-2.0	12B	256K	-	Jun 3, 2026	source
Step-3.7-Flash	StepFun	Apache-2.0	196B · 11B active	256K	-	May 29, 2026	source
LFM2.5-8B-A1B	Liquid	LFM Open License v1.0	8.3B · 1.5B active	131K	-	May 28, 2026	source
MiniMax-M2.7	MiniMax	MiniMax Model License	229.9B · 9.8B active	—	-	May 26, 2026	source
DeepSeek V4-Flash	DeepSeek	MIT	284B · 13B active	1M	-	Apr 24, 2026	source
DeepSeek V4-Pro	DeepSeek	MIT	1.6T · 49B active	1M	-	Apr 24, 2026	source
Hunyuan Hy3-preview	Hunyuan	Tencent License	295B · 21B active	256K	-	Apr 23, 2026	source
Hunyuan-A13B-Instruct	Hunyuan	Tencent Hunyuan A13B License	80B · 13B active	—	-	Apr 22, 2026	source
MiMo-V2.5-Pro	Xiaomi	MIT	1.0T · 42B active	1M	-	Apr 22, 2026	source
MiMo-V2.5	Xiaomi	MIT	310B · 15B active	1M	-	Apr 22, 2026	source
GLM-5.1	Z.ai	MIT	754B · 40B active	—	-	Apr 8, 2026	source
Gemma 4 31B	DeepMind	Apache-2.0	31B	—	-	Apr 2, 2026	source
Nemotron 3 Super 120B-A12B	NVIDIA	Nemotron Open Model License	120B · 12B active	1M	-	Mar 16, 2026	source
Mistral Small 4	Mistral	Apache 2.0	119B · 6B active	256K	-	Mar 16, 2026	source
Step-3.5-Flash	StepFun	Apache-2.0	196B · 11B active	256K	-	Mar 14, 2026	source
Qwen3.5-9B	Qwen	Apache-2.0	9B	262K	-	Mar 2, 2026	source
Qwen3.5-4B	Qwen	Apache-2.0	4B	262K	-	Mar 2, 2026	source
Qwen3.5-2B	Qwen	Apache-2.0	2B	262K	-	Mar 2, 2026	source
Qwen3.5-0.8B	Qwen	Apache-2.0	0.8B	262K	-	Mar 2, 2026	source
Qwen3.5-397B	Qwen	Apache-2.0	397B · 17B active	1M	-	Feb 20, 2026	source
OLMo 3 Think 32B	Ai2	Apache-2.0	32B	—	-	Dec 15, 2025	source
Nemotron 3 Nano 30B-A3B	NVIDIA	Nemotron Open Model License	30B · 3B active	1M	-	Dec 15, 2025	source
GLM-4.6V	Z.ai	MIT	106B	128K	-	Dec 8, 2025	source
Mistral Large 3	Mistral	Mistral Research / Commercial	675B · 41B active	256K	-	Dec 2, 2025	source
DeepSeek-V3.2-Speciale	DeepSeek	MIT	685B · 37B active	128K	-	Dec 1, 2025	source
LFM2 1.2B	Liquid	LFM Open License v1.0	1.17B	33K	-	Nov 28, 2025	source
Kimi-Linear-48B-A3B-Instruct	Moonshot	MIT	48B · 3B active	1.0M	-	Oct 31, 2025	source
GLM-4.6	Z.ai	MIT	357B · 32B active	200K	-	Sep 30, 2025	source
DeepSeek-V3.2-Exp	DeepSeek	MIT	685B · 37B active	128K	-	Sep 29, 2025	source
DeepSeek-V3.1	DeepSeek	MIT	671B · 37B active	128K	-	Aug 21, 2025	source
DeepSeek R2	DeepSeek	Expected MIT (unconfirmed)	Undisclosed	—	-	Aug 14, 2025	source
gpt-oss-20b	OpenAI	Apache-2.0	21B · 3.6B active	128K	-	Aug 5, 2025	source
gpt-oss-120b	OpenAI	Apache-2.0	117B · 5.1B active	128K	-	Aug 5, 2025	source
Falcon-H1 34B	TII	Falcon LLM License 2.0	34B	256K	-	Jul 31, 2025	source
GLM-4.5	Z.ai	MIT	355B · 32B active	128K	-	Jul 28, 2025	source
GLM-4.5-Air	Z.ai	MIT	106B · 12B active	128K	-	Jul 28, 2025	source
Qwen3-Coder-480B-A35B-Instruct	Qwen	Apache-2.0	480B · 35B active	262K	-	Jul 22, 2025	source
EXAONE 4.0 32B	LG AI	EXAONE AI Model License	32B	—	-	Jul 15, 2025	source
SmolLM3 3B	Hugging Face	Apache-2.0	3B	128K	-	Jul 8, 2025	source
ERNIE-4.5-300B-A47B	Baidu	Apache-2.0	300B · 47B active	128K	-	Jun 30, 2025	source
Magistral Small	Mistral	Apache-2.0	24B	40K	-	Jun 10, 2025	source
Sarvam-M	Sarvam	Sarvam AI License	Undisclosed	—	-	May 21, 2025	source
Devstral Small 2505	Mistral	Apache-2.0	24B	128K	-	May 21, 2025	source
Gemma 3n E4B	DeepMind	Gemma Terms of Use	8B · 4B active	32K	-	May 20, 2025	source
Phi-4 Reasoning	Microsoft	MIT	14B	—	-	Apr 30, 2025	source
Granite 3.3 8B	IBM	Apache-2.0	8B	128K	-	Apr 30, 2025	source

Frequently asked questions

What makes a model an “open coding model”?

It must be downloadable (open-weight or open-source) and positioned for software work — coding, repository tasks, tool use, or software-engineering agents. Models behind a closed API are covered in the frontier leaderboard instead.

Is SWE-bench the only thing that matters for coding?

No. SWE-bench Verified is the most comparable single number, but context window (how much of a repo fits), license terms for commercial use, and inference cost or local hardware fit often decide which model is actually usable for you.

Can I use these models commercially?

Often, but check each license. Open-source licences like Apache-2.0 and MIT are permissive; many open-weight models carry custom terms with scale, competitor, or redistribution restrictions. See the open-weight guide for a checklist.