Which model is the best coding agent?

Rankings

Coding agents and models, ranked. Hold the agent harness constant, swap only the model underneath, and score the result.

Aggregated across 9 public benchmarks — CodingAgentBench, SWE-bench, Terminal-Bench, Chatbot Arena, Arena Agent, ixio runs, Stratix Cup — into one composite, percentile-blended so models that actually ship rank highest. Live model runs go through Nebius Token Factory.

● Live45 agents328 modelsUpdated Jun 26, 2026

ixio.com View on GitHub

Top models

Models ranked by a composite blended across every benchmark they appear on. Showing the top 60 of 328 — full set on ixio.

#	Model	Vendor	Score	Tier	Best result
1	Grok 4.3 xAI	Unknown	100.0	Excellent	Arena Agent · 17.72
2	Nemotron-3 Ultra 550BOpen	NVIDIA	100.0	Excellent	ixio runs · 100
3	Claude Opus 4.8	Anthropic	98.2	Excellent	SWE-bench Verified · 88.6
4	GPT-5.5	OpenAI	96.4	Excellent	Terminal-Bench 2.0 · 84.7
5	Gemma 4 31B	Google	96.3	Excellent	Arena Agent · 13.01
6	ERNIE-5.1	Baidu	93.3	Excellent	Artificial Analysis · 71
7	Qwen3.7-Max	Alibaba	92.9	Excellent	Artificial Analysis · 72
8	Muse Spark	Meta	92.9	Excellent	Artificial Analysis · 71
9	Kimi-K2.6-Thinking ✅Open	Moonshot AI	92.9	Excellent	Artificial Analysis · 71
10	DeepSeek-V4-Pro ✅Open	DeepSeek	92.4	Excellent	Artificial Analysis · 71
11	Seed2.0 Pro	ByteDance	91.3	Excellent	Chatbot Arena (Coding) · 1495
12	Qwen3.5-Max	Alibaba	91.0	Excellent	Chatbot Arena (Coding) · 1493
13	mistral-small-4-119b-2603Open	Mistral AI	90.7	Excellent	CodingAgentBench · 80
14	Qwen3.6-Plus	Alibaba	89.6	Excellent	Chatbot Arena (Coding) · 1482
15	nemotron-3-ultra-550b-a55bOpen	NVIDIA	89.5	Excellent	CodingAgentBench · 92
16	Claude Opus 4.7	Anthropic	89.1	Excellent	Artificial Analysis · 76
17	Gemini 3.1 Pro	Google	88.2	Excellent	Artificial Analysis · 76
18	Claude Opus 4.6	Anthropic	87.7	Excellent	Chatbot Arena (Coding) · 1545
19	Grok 4.3	xAI	87.5	Excellent	Stratix Cup · 51.1
20	step-3.5-flashOpen	StepFun	86.0	Excellent	CodingAgentBench · 92
21	GPT-5.3-Codex	OpenAI	85.6	Excellent	Terminal-Bench 2.0 · 78.4
22	Claude Opus 4.5 (thinking-32k)	Anthropic	85.6	Excellent	Chatbot Arena (Coding) · 1510
23	Grok 4.20	xAI	85.4	Excellent	Chatbot Arena (Coding) · 1518
24	Gemini 3 Pro	Google	85.2	Excellent	Artificial Analysis · 73
25	MiMo-V2.5-ProOpen	Xiaomi	85.2	Excellent	Chatbot Arena (Coding) · 1485
26	GLM-5.2Open	Z.ai	84.4	Excellent	ixio runs · 100
27	Gemini 3 Flash	Google	84.2	Excellent	Artificial Analysis · 71
28	Kimi K2.7Open	Moonshot AI	83.9	Excellent	SWE-bench Verified · 78.2
29	Kimi K2.6Open	Moonshot AI	83.9	Excellent	ixio runs · 100
30	Grok-4.1	xAI	83.8	Excellent	Chatbot Arena (Coding) · 1463
31	GPT-5.2-Codex	OpenAI	83.6	Excellent	Terminal-Bench 2.0 · 66.5
32	Grok-4.1-Thinking	xAI	82.1	Excellent	Chatbot Arena (Coding) · 1483
33	GPT-5.2	OpenAI	81.8	Excellent	Artificial Analysis · 72
34	Kimi K2.7 CodeOpen	Moonshot AI	81.3	Excellent	Stratix Cup · 41.5
35	ERNIE-5.0	Baidu	80.9	Excellent	Chatbot Arena (Coding) · 1461
36	MiMo-V2-Pro	Xiaomi	80.5	Excellent	Artificial Analysis · 69
37	Kimi-K2-ThinkingOpen	Moonshot AI	79.9	Excellent	Artificial Analysis · 67
38	Gemma-4-31B-itOpen	Google	79.2	Excellent	Chatbot Arena (Coding) · 1462
39	Hunyuan-Hy3	Tencent	79.1	Excellent	Artificial Analysis · 66
40	MiniMax M3Open	MiniMax	78.4	Excellent	Chatbot Arena (Coding) · 1485
41	MiniMax M2.7Open	MiniMax	78.2	Excellent	CodingAgentBench · 84
42	Claude Opus 4.1 (thinking-16k)	Anthropic	78.2	Excellent	Chatbot Arena (Coding) · 1475
43	GLM-5.1Open	Z.ai	77.1	Excellent	Chatbot Arena (Coding) · 1506
44	qwen3-next-80b-a3b-instructOpen	Alibaba	76.6	Excellent	CodingAgentBench · 68
45	GPT-5.4-mini	OpenAI	76.6	Excellent	Chatbot Arena (Coding) · 1461
46	Kimi-K2.5-ThinkingOpen	Moonshot AI	76.5	Excellent	Chatbot Arena (Coding) · 1480
47	nemotron-3-super-120b-a12bOpen	NVIDIA	76.3	Excellent	CodingAgentBench · 84
48	Qwen-VL-Max-2025-08-13	Alibaba	75.9	Excellent	Chatbot Arena (Coding) · 1440
49	DeepSeek-V4-FlashOpen	DeepSeek	75.6	Excellent	Chatbot Arena (Coding) · 1460
50	GPT-5.4	OpenAI	75.4	Excellent	Chatbot Arena (Coding) · 1538
51	Qwen3-235B-A22B-Thinking-2507Open	Alibaba	74.7	Excellent	Chatbot Arena (Coding) · 1442
52	GPT-5.1-Codex-Mini	OpenAI	74.6	Excellent	Terminal-Bench 2.0 · 61.6
53	Claude Sonnet 4.6	Anthropic	74.5	Excellent	Chatbot Arena (Coding) · 1511
54	Grok-4-Fast	xAI	74.5	Excellent	Chatbot Arena (Coding) · 1441
55	MiniMax M2.7 MiniMax · Modified	MIT	74.1	Excellent	Arena Agent · 7.94
56	Gemma-4-26B-A4B-itOpen	Google	73.5	Excellent	Chatbot Arena (Coding) · 1460
57	Claude Opus 4.5	Anthropic	73.0	Excellent	Chatbot Arena (Coding) · 1496
58	Claude Sonnet 4.5 (thinking-32k)	Anthropic	72.9	Excellent	Chatbot Arena (Coding) · 1485
59	Qwen3-VL-235B-A22B-ThinkingOpen	Alibaba	72.2	Excellent	Chatbot Arena (Coding) · 1432
60	Amazon-Nova-Chat-11-10	Amazon	72.1	Excellent	Artificial Analysis · 62

Top open-source coding agents

Open-source harnesses, ranked using only their open-weight models — the 20 of 45 agents with an open-model result. Best model is the strongest open weight each one ran.

#	Agent	Score	Tier	Best model	Models
1	AiderCLI	91.3	Excellent	gpt-oss-120b	14
2	OpenCodeTUI	86.7	Excellent	gpt-oss-120b	14
3	Qwen CodeCLI	84.6	Excellent	step-3.5-flash	14
4	mini-SWE-agentAgent	84.3	Excellent	GLM-5.2	14
5	Claude CodeCLI	84.0	Excellent	GLM-5.2	10
6	CrushTUI	82.6	Solid	nemotron-3-super-120b-a12b	14
7	Copilot CLICLI	79.9	Solid	step-3.5-flash	14
8	piCLI	75.0	Solid	gpt-oss-120b	14
9	Codex CLICLI	71.1	Solid	nemotron-3-super-120b-a12b	14
10	OpenHandsAgent	71.1	Solid	step-3.5-flash	24
11	GooseCLI	65.2	Solid	gpt-oss-120b	14
12	PlandexCLI	59.5	Solid	gpt-oss-120b	14
13	IndusAGI Coding AgentAgent	54.5	Iffy	MiniMax M2.7	1
14	Harness AgentAgent	49.3	Iffy	MiniMax M2.7 Highspeed	1
15	Terminus 2Agent	48.7	Iffy	GLM-5	14
16	cchuterAgent	47.8	Iffy	MiniMax M2.5	1
17	CruxAgent	38.2	Iffy	MiniMax M2.1	2
18	Dakou AgentAgent	25.4	Iffy	Qwen 3 Coder 480B	1
19	little-coderAgent	16.8	Iffy	Qwen3.6-35B-A3B	2
20	Bash AgentAgent	13.4	Iffy	TermiGen-32B	1

Aggregated from public leaderboards. Each cell is that benchmark's headline metric; composites are percentile-blended across benchmarks. Data from the open-source ensemble leaderboard, 935 entries scraped Jun 26, 2026.