EvoCode-Bench - Task Completion: leaderboard

EvoCode-Bench (2026) full-task completion: the percent of 26 multi-round coding tasks (5 to 15 rounds of evolving and conflicting requirements in one persistent workspace) that an agent carries through the final round without a failed cumulative check, in any of four attempts, each model in the Terminus-2 harness; higher is better.

Metric: Full-task completion (Comp, %): share of the 26 tasks in which at least one of four attempts passes the cumulative verifier at every round through the final one (fail-stop), over the 26 stateful multi-round coding tasks of EvoCode-Bench (227 rounds, 5 to 15 per task, requirements that extend, correct or contradict earlier ones, cumulative executable verifiers), each model driving the Terminus-2 agent harness in a Harbor Docker workspace; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 13 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Opus 4.7High42.3#85
2GPT-5.5High38.5#54
3Claude Opus 4.634.6#113
4Kimi K2.623.1#192
5DeepSeek V4 ProHigh19.2#196
6GLM-5.115.4#198
7Qwen 3.6 PlusThinking15.4#238
8Gemini 3.1 Pro (Preview)High11.5#99
9MiMo-V2.5-ProHigh11.5#284
10Qwen 3.5 397B A17B0#264
11MiniMax-M2.70#539
12DeepSeek V4 FlashHigh0#267
13Seed 2.0 ProHigh0#222

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=evocode-bench-task-completion · How It Works · Data refreshed daily, snapshot 2026-11-21.