What changed this week
Every leaderboard move, new model and price change, week by week.
Week of 2026-10-13
- New modelClaude Haiku 5.5 entered the board at #14 with an 87.9 AMR score — the largest small-model generation jump of the year.
- New modelClaude Mythos (Preview) is now tracked at a provisional 91.0 AMR score pending public release.
- Rank changeGemini 4 Argon climbed 2 spots to #7 after fresh LiveCodeBench and arena data landed; GLM-5.3 Flash jumped 3 spots to #21.
- Rank changeGPT-6.1 Sol overtook GPT-6 Astra for #2 on the back of updated AIME 2025 results (97.2%).
- NoteSnapshot date moved to 2026-10-14. All prices re-verified against vendor rate cards.
Week of 2026-10-06
- New modelClef 2 joined the open tier with a focus on structured JSON output; initial Elo 1259.
- Benchmark updateSWE-Bench Verified re-run for the full Claude 5.x family; Opus 5.5 confirmed at 81.4%.
- Price changeGemini 3.1 Pro output price held at $5.00/M — third week stable; DeepSeek Harness remains the price-performance leader.
Week of 2026-09-29
Updates FAQ
How often does the leaderboard change?
The leaderboard is recomputed weekly and new models are added within 48 hours of release. In October 2026 the board moved three times in three weeks — new entries like Claude Haiku 5.5, Grok 5 Preview and DeepSeek Harness each shifted several ranks on arrival.