Skip to content
  1. Home
  2. Releases
rev 01 · update

Four models added; Claude Opus 4.8 leaves the board

Tencent HY 4 Preview, GLM 5.3 Flash, Muse Spark 1.3 and GLM 5.3 join the public leaderboard. Claude Opus 4.8 is retired. The board is 20 models; 200 new attempt bundles are published.

5 Sep 2026leaderboard

What shipped

  • Tencent HY 4 Preview via OpenCode at max: 62.62, 32 of 50 resolved, fifth place. Model page.
  • GLM 5.3 Flash via OpenCode: 59.37, 30 of 50 resolved, ninth. Model page.
  • Muse Spark 1.3 via OpenCode: 57.66, 29 of 50 resolved, twelfth. Model page.
  • GLM 5.3 via OpenCode: 57.31, 29 of 50 resolved, fourteenth. Model page.
  • Data. 200 new attempt bundles at data.integrationbench.com, same layout as the rest. Leaderboard, matrix, per-task indexes and trajectory pages include the four models.
  • Claude Opus 4.8 (55.58) leaves the leaderboard, matrix, per-task pages and published attempt set. Same treatment as Haiku 4.5 and Gemini 3.7 Flash. The rev 01 launch note still records the original score.

Notes

  • Same task set, same scorer (task-score-v4-mandatory-gated), one official attempt per task.
  • Existing published scores do not change. Ranks below the new entries shift.
  • The board is 20 models, 1,000 published attempts.