Coding · sub-raceNarrow lead
Agentic Coding race
Autonomous, multi-hour software engineering in large repositories: planning, editing many files, running tests and iterating.
- Leader
- Sonnet 5.5
- Margin
- 0.8 pts
- 0.9% ahead
- Weeks at #1
- 4
- in a row
- Changes
- 6
- in 2026
Warming up the track…
Oct 19, 2026Sonnet 5.5 leads
2025Apr ’25Jul ’25Oct ’252026Apr ’26Jul ’26Oct ’26
Score history
Top 6 today over the leader's frontier · click to jump to a week
#1 frontierSonnet 5.5Opus 5.5Fable 5.1GPT-6 AstraGPT-6.1 SolGrok 4.7
Reigns at #1
Longest: Opus 4, 11 weeks (from May 2025).
Sibling races
More coding races
Why this race matters
This is the capability behind Claude Code, Codex and Cursor agents, and the fastest-moving race on the site.
How it is scored
Resolved-task rate with a standard agent harness and fixed tool budget.
Unit: percentHigher is betterPhoto finish under 0.5 ptsDominant over 3.0 pts
Data provenance
Contamination-resistant, long-horizon engineering tasks.
DeepSWE
Multi-file agent tasks scored by hidden tests.
Scores here are illustrative placeholders that mimic real benchmark patterns. Swap in live data through the score provider. Methodology