OpenCode + Kimi K2.5: Open Source, Open Weight Agentic Coding is Here

OpenCode + Kimi K2.5 brings near-frontier agentic coding performance at a fraction of the cost and at much higher speed.
As of 29 January 2026
Latest Leaderboard
We compare the latest on , , , and , grouping them by relative performance for each metric.
Alerts
Agent
History
Sigmascore
Accuracy
Consistency
Speed
Cursor CLI
Composer-1
42.9%
39.1%
50.9%
39.5%
Cursor CLI
new
Composer-1.5
38.3%
38.6%
51.4%
28.3%
OpenCode
Gemini 3 Flash Preview
37.7%
40.6%
57.2%
23.1%
Cursor CLI
Grok Code Fast 1
37.6%
34.4%
47.6%
32.4%
OpenCode
Gemini 3 Pro Preview
33.0%
38.1%
51.2%
18.4%
OpenCode
new
Kimi K2.5
32.7%
39.4%
46.8%
18.9%
Codex CLI
GPT-5.2-Codex
32.4%
45.9%
50.3%
14.8%
Gemini CLI
Gemini 3 Flash Preview
32.4%
38.0%
50.2%
17.9%
Claude Code
Claude 4.5 Opus
32.0%
43.1%
49.8%
15.3%
Claude Code
new
Claude 4.6 Opus
31.3%
47.6%
52.1%
12.4%
Codex CLI
GPT-5.1-Codex-Max
30.2%
44.3%
49.9%
12.5%
OpenCode
GPT-5.2 Codex
29.6%
43.3%
45.9%
13.1%
Gemini CLI
Gemini 3 Pro Preview
28.9%
36.0%
46.0%
14.6%
Codex CLI
GPT-5.1-Codex-Mini
27.5%
40.0%
42.6%
12.2%
OpenCode
GPT-5.1 Codex Mini
25.1%
38.9%
42.9%
9.4%
OpenCode
new
MiniMax M2.5
25.1%
33.8%
34.6%
13.5%
Claude Code
Claude Haiku 4.5
24.6%
26.2%
32.8%
17.2%
Codex CLI
GPT-5.1-Codex
23.4%
40.2%
40.3%
7.9%
Gemini CLI
Gemini 2.5 Flash
21.5%
17.9%
30.6%
18.1%
Warp
—
Request
—
—
—
 
Amp
—
Request
—
—
—
 
Mistral Vibe CLI
—
Request
—
—
—
 
Cursor CLI
Composer-1
Accuracy:
39.1%
Consistency:
50.9%
Speed:
39.5%
42.9%
Cursor CLI
new
Composer-1.5
Accuracy:
38.6%
Consistency:
51.4%
Speed:
28.3%
38.3%
OpenCode
Gemini 3 Flash Preview
Accuracy:
40.6%
Consistency:
57.2%
Speed:
23.1%
37.7%
Cursor CLI
Grok Code Fast 1
Accuracy:
34.4%
Consistency:
47.6%
Speed:
32.4%
37.6%
OpenCode
Gemini 3 Pro Preview
Accuracy:
38.1%
Consistency:
51.2%
Speed:
18.4%
33.0%
OpenCode
new
Kimi K2.5
Accuracy:
39.4%
Consistency:
46.8%
Speed:
18.9%
32.7%
Codex CLI
GPT-5.2-Codex
Accuracy:
45.9%
Consistency:
50.3%
Speed:
14.8%
32.4%
Gemini CLI
Gemini 3 Flash Preview
Accuracy:
38.0%
Consistency:
50.2%
Speed:
17.9%
32.4%
Claude Code
Claude 4.5 Opus
Accuracy:
43.1%
Consistency:
49.8%
Speed:
15.3%
32.0%
Claude Code
new
Claude 4.6 Opus
Accuracy:
47.6%
Consistency:
52.1%
Speed:
12.4%
31.3%
Codex CLI
GPT-5.1-Codex-Max
Accuracy:
44.3%
Consistency:
49.9%
Speed:
12.5%
30.2%
OpenCode
GPT-5.2 Codex
Accuracy:
43.3%
Consistency:
45.9%
Speed:
13.1%
29.6%
Gemini CLI
Gemini 3 Pro Preview
Accuracy:
36.0%
Consistency:
46.0%
Speed:
14.6%
28.9%
Codex CLI
GPT-5.1-Codex-Mini
Accuracy:
40.0%
Consistency:
42.6%
Speed:
12.2%
27.5%
OpenCode
GPT-5.1 Codex Mini
Accuracy:
38.9%
Consistency:
42.9%
Speed:
9.4%
25.1%
OpenCode
new
MiniMax M2.5
Accuracy:
33.8%
Consistency:
34.6%
Speed:
13.5%
25.1%
Claude Code
Claude Haiku 4.5
Accuracy:
26.2%
Consistency:
32.8%
Speed:
17.2%
24.6%
Codex CLI
GPT-5.1-Codex
Accuracy:
40.2%
Consistency:
40.3%
Speed:
7.9%
23.4%
Gemini CLI
Gemini 2.5 Flash
Accuracy:
17.9%
Consistency:
30.6%
Speed:
18.1%
21.5%
Warp
Amp
Mistral Vibe CLI
Why the best coding agent depends on your codebase.
Across the Sigmabench dataset, coding agent performance varies widely between similar codebases. These differences emerge from the unique characteristics of each codebase — shaped over time by the humans who built it — and cannot be reliably inferred from large-scale features like programming language, problem domain, or codebase size. This makes common decisions hard to answer with confidence:
Which agent should we standardize on?
Are we using the right agent for our codebase?
Would switching agents materially improve accuracy or speed?
Are recent improvements real, or just hype?
Sigmabench lets you run the same benchmark used for the public leaderboard on your own codebase, so you can answer these questions with data — measuring accuracy, consistency, and speed in your real-world development environment.
Benchmarks are read-only and SOC 2-compliant.