Leaderboard
We select harnesses and inference endpoints to bring out each model’s best performance.
Ranked by task pass rate, with mean cost and median runtime reported per task. A task passes only when all its cases pass.
| Model | Harness | Pass rate | Tokens | Cost | Time | Cache | Per-task result |
|---|---|---|---|---|---|---|---|
1 GPT-6 Astra | Codex | 34.1% 15/44 | 2.3M | $3.48 | 11m 48s | 97% | |
2 Claude Opus 5.5 | Claude Code | 27.3% 12/44 | 10.9M | $3.97 | 20m 50s | 98% | |
3 DeepSeek Flash open | DSH | 15.9% 7/44 | 29.0M | $0.23 | 23m 56s | 99% | |
4 GLM-5.3 Flash open | ZCode | 9.1% 4/44 | 24.8M | $0.82 | 1h 02m | 99% | |
5 Qwen3.8 Max | Codex | 6.8% 3/44 | 16.6M | $5.12 | 1h 04m | 98% | |
6 Kimi K3 open | Kimi Code | 4.5% 2/44 | 18.1M | $7.92 | 1h 01m | 98% |
Compare task pass rates with mean cost and median runtime across all tasks, including failures.
GPT-6 AstraClaude Opus 5.5DeepSeek FlashGLM-5.3 FlashQwen3.8 MaxKimi K3
Cost includes subagents and is averaged over all 44 tasks. Time is the median per task run.
GPT-6 Astra
Actual Budget0/1
Afilmory1/2
Cal.diy0/1
Dify0/1
Discourse0/2
Documenso1/1
Dolibarr1/5
FreshRSS0/1
GLPI1/2
Gitea1/4
Grist1/1
Hoppscotch1/1
Kimai0/1
LimeSurvey0/1
LobeHub0/1
Metabase0/3
Navidrome0/1
NetBox1/2
OpenEMR1/1
Oppia1/2
Penpot1/3
Plane1/3
Saleor Dashboard2/2
Snipe-IT1/1
ownCloud Infinite Scale0/1
Claude Opus 5.5
Actual Budget0/1
Afilmory1/2
Cal.diy0/1
Dify0/1
Discourse0/2
Documenso0/1
Dolibarr1/5
FreshRSS0/1
GLPI1/2
Gitea2/4
Grist0/1
Hoppscotch0/1
Kimai1/1
LimeSurvey0/1
LobeHub0/1
Metabase0/3
Navidrome1/1
NetBox1/2
OpenEMR1/1
Oppia1/2
Penpot1/3
Plane1/3
Saleor Dashboard0/2
Snipe-IT0/1
ownCloud Infinite Scale0/1
DeepSeek Flash
Actual Budget0/1
Afilmory1/2
Cal.diy0/1
Dify0/1
Discourse0/2
Documenso0/1
Dolibarr0/5
FreshRSS0/1
GLPI0/2
Gitea0/4
Grist1/1
Hoppscotch0/1
Kimai1/1
LimeSurvey0/1
LobeHub0/1
Metabase0/3
Navidrome1/1
NetBox0/2
OpenEMR0/1
Oppia0/2
Penpot1/3
Plane0/3
Saleor Dashboard1/2
Snipe-IT1/1
ownCloud Infinite Scale0/1
GLM-5.3 Flash
Actual Budget0/1
Afilmory0/2
Cal.diy0/1
Dify0/1
Discourse0/2
Documenso0/1
Dolibarr0/5
FreshRSS0/1
GLPI0/2
Gitea1/4
Grist0/1
Hoppscotch0/1
Kimai1/1
LimeSurvey0/1
LobeHub1/1
Metabase0/3
Navidrome0/1
NetBox0/2
OpenEMR0/1
Oppia0/2
Penpot0/3
Plane0/3
Saleor Dashboard1/2
Snipe-IT0/1
ownCloud Infinite Scale0/1
Qwen3.8 Max
Actual Budget0/1
Afilmory0/2
Cal.diy0/1
Dify0/1
Discourse0/2
Documenso0/1
Dolibarr1/5
FreshRSS0/1
GLPI1/2
Gitea0/4
Grist0/1
Hoppscotch0/1
Kimai0/1
LimeSurvey0/1
LobeHub0/1
Metabase0/3
Navidrome0/1
NetBox0/2
OpenEMR0/1
Oppia0/2
Penpot0/3
Plane0/3
Saleor Dashboard1/2
Snipe-IT0/1
ownCloud Infinite Scale0/1
Kimi K3
Actual Budget0/1
Afilmory1/2
Cal.diy0/1
Dify0/1
Discourse0/2
Documenso0/1
Dolibarr0/5
FreshRSS0/1
GLPI0/2
Gitea0/4
Grist0/1
Hoppscotch0/1
Kimai0/1
LimeSurvey0/1
LobeHub0/1
Metabase0/3
Navidrome0/1
NetBox0/2
OpenEMR0/1
Oppia0/2
Penpot0/3
Plane0/3
Saleor Dashboard1/2
Snipe-IT0/1
ownCloud Infinite Scale0/1