Project benchmark

CompassBench 2026-07 · Code

OpenCompass · 15 configurations with published scores. Review each metric and its test conditions separately.

Report identity & conditions

Data as of2026-10-19· Report published—. See individual records for run dates and test conditions.

Leaderboard updated2026-08-24

This leaderboard uses OpenCompass’s closed benchmark sets. Questions update quarterly, and models may be added between quarters. Quarters and tasks are compared separately.
Overall and task scores follow the original leaderboard. We do not recalculate overall weights or interpret scores as question accuracy.
The same model configurations appear across five task scopes. Reasoning settings stay in the names; adding row counts does not give a count of unique models.
Organizations, parameter counts, release dates and release types follow the report. An open-weight label does not imply an open license; use depends on the model’s terms.
Quarterly versions and leaderboard update dates are kept separate. Actual test dates and individual sample counts are not stated.
Results: official OpenCompass leaderboard.
Quarterly version: 2026-07; leaderboard updated: 2026-08-24.

Left metricRight metric

Compare metrics within the same report. Each chart explains its direction and scope.
See the details for full values and run conditions.

Code score

Original scores from this quarter’s closed benchmark; higher is better. Overall and task values are retained and are not ranked across quarters or tasks.: showing 8 of 15 records with values. See details for configurations and dates.

Chart data · 8 entries

Name	points
GPT-5.5 (high)	96.5
GPT-5.6-Sol (high)	96.3
Qwen3.8-Max	95.5
Claude Opus 5 (high)	93.3
DeepSeek-V4-Pro	91
Doubao-Seed-2.1-Pro-260628 (high)	90.8
Gemini-3.7-Flash	89.3
Kimi-K3 (high)	88

Coding, including agents score

Chart data · 8 entries

Name	points
GPT-5.5 (high)	96.5
GPT-5.6-Sol (high)	96.3
Qwen3.8-Max	95.5
Claude Opus 5 (high)	93.3
DeepSeek-V4-Pro	91
Doubao-Seed-2.1-Pro-260628 (high)	90.8
Gemini-3.7-Flash	89.3
Kimi-K3 (high)	88

Full results & model versions

15 / 15configurations have published scores in this scope. Scored configurations are listed; missing values remain “Not provided.” Different run configurations of the same model are listed separately.

Gemini-3.7-Flash

Model reported by the benchmark

As reportedOrganization: Google · Model details ↗

Chat

Release date reported by the benchmark:2026-08-13

API

Code score: 89.3 分
Coding, including agents score: 89.3 分

GLM-5.3

Model reported by the benchmark

As reportedOrganization: Zhipu AI · Model details ↗

Chat

Release date reported by the benchmark:2026-08-14

API

Code score: 76.5 分
Coding, including agents score: 76.5 分

Kimi-K3 (high)

Model reported by the benchmark

As reportedOrganization: Moonshot · Model details ↗

Chat

Release date reported by the benchmark:2026-07-16

Parameters: 2.8T · Open weights

Code score: 88 分
Coding, including agents score: 88 分

Hy3-preview (high)

Model reported by the benchmark

As reportedOrganization: Tencent · Model details ↗

Chat

Release date reported by the benchmark:2026-04-23

Parameters: 295B · Open weights

Code score: 82 分
Coding, including agents score: 82 分

Doubao-Seed-2.1-Pro-260628 (high)

Model reported by the benchmark

As reportedOrganization: ByteDance · Model details ↗

Chat

Release date reported by the benchmark:2026-06-28

API

Code score: 90.8 分
Coding, including agents score: 90.8 分

GPT-5.5 (high)

Model reported by the benchmark

As reportedOrganization: OpenAI · Model details ↗

Chat

Release date reported by the benchmark:2026-04-23

API

Code score: 96.5 分
Coding, including agents score: 96.5 分

GPT-5.6-Sol (high)

Model reported by the benchmark

As reportedOrganization: OpenAI · Model details ↗

Chat

Release date reported by the benchmark:2026-07-09

API

Code score: 96.3 分
Coding, including agents score: 96.3 分

Kimi-K2.6 (high)

Model reported by the benchmark

As reportedOrganization: Moonshot · Model details ↗

Chat

Release date reported by the benchmark:2026-04-20

Parameters: 1T · Open weights

Code score: 74.5 分
Coding, including agents score: 74.5 分

Qwen3.8-Max

Model reported by the benchmark

As reportedOrganization: Alibaba · Model details ↗

Chat

Release date reported by the benchmark:2026-08-02

Parameters: 2.4T · Open weights

Code score: 95.5 分
Coding, including agents score: 95.5 分

DeepSeek-V4-Pro

Model reported by the benchmark

As reportedOrganization: DeepSeek · Model details ↗

Chat

Release date reported by the benchmark:2026-04-24

Parameters: 1.6T · Open weights

Code score: 91 分
Coding, including agents score: 91 分

Grok-4.6 (high)

Model reported by the benchmark

As reportedOrganization: xAI · Model details ↗

Chat

Release date reported by the benchmark:2026-08-12

API

Code score: 85 分
Coding, including agents score: 85 分

DeepSeek-V4-Flash

Model reported by the benchmark

As reportedOrganization: DeepSeek · Model details ↗

Chat

Release date reported by the benchmark:2026-04-24

Parameters: 284B · Open weights

Code score: 79 分
Coding, including agents score: 79 分

Claude Opus 4.8 (high)

Model reported by the benchmark

As reportedOrganization: Anthropic · Model details ↗

Chat

Release date reported by the benchmark:2026-05-28

API

Code score: 77.3 分
Coding, including agents score: 77.3 分

MiniMax-M3

Model reported by the benchmark

As reportedOrganization: MiniMax · Model details ↗

Chat

Release date reported by the benchmark:2026-06-01

Parameters: 428B · Open weights

Code score: 72.5 分
Coding, including agents score: 72.5 分

Claude Opus 5 (high)

Model reported by the benchmark

As reportedOrganization: Anthropic · Model details ↗

Chat

Release date reported by the benchmark:2026-07-24

API

Code score: 93.3 分
Coding, including agents score: 93.3 分