IT之家 · ITHome
Intelligence Index v5 launches; DeepSeek leads released-weight models on a coding test
Artificial Analysis has launched Intelligence Index v5, putting 60% of its evaluation weight on private tests or hidden answers to limit benchmark contamination. DeepSeek V4.1-Flash 1017 (max) scored 28% on FrontierCode 2.0, leading models with released weights. Its overall index score was 32, versus the leader’s 58. The coding test asks repository maintainers whether submitted changes merit merging.
What to know
- 01
FrontierCode 2.0 accounts for 15% of the index and covers 132 tasks from real open-source repositories; any failed blocking check results in zero for that task.
- 02
DeepSeek V4.1-Flash 1017 (max) scored 28% on this test. In the overall index, DeepSeek and GLM-5.3 each scored 32, while leader Claude Opus 5.5 scored 58.
Why it matters
Because the index covers multiple capability areas, leadership on FrontierCode 2.0 should not be read as leadership in the overall ranking.
AI-assisted brief. Read the source for full details. Editorial & sourcing policy
