LLM Stats (Toolathlon - Verified protocol): leaderboard
Vendor-reported Toolathlon-Verified runs explicitly identified in the per-row methods of the original LLM Stats Toolathlon collection. The June 30 Verified release changed tasks and graders. Preserve source-specific harnesses and all values; keep the separately collected Toolathlon-Verified board distinct until exact protocol equivalence is established. Score percentage; higher is better.
Metric: Score (%). Source: llm-stats.com. Saturation forecast: Around January 2027. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5.3 Flash | 78.4 | #65 |
| 2 | Muse Spark 1.1 | 75.6 | #52 |
| 3 | DeepSeek V4 Pro (0813) | 74.1 | #53 |
| 4 | Hy4 preview | 74.1 | #38 |
| 5 | Qwen3.8-Flash-Next | 73.5 | #72 |
| 6 | Qwen 3.8 Flash | 73.5 | #50 |
| 7 | Kimi K3 | 73.2 | #30 |
| 8 | GLM-5.3 | 73 | #39 |
| 9 | Qwen 3.8 Max | 72.5 | #35 |
| 10 | DeepSeek V4 Flash (0731) | 70.3 | #78 |
| 11 | Inkling Small | 54.4 | #138 |
| 12 | Laguna S 2.1 | 49.7 | #462 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=llm-stats-toolathlon-verified-protocol · How It Works · Data refreshed daily, snapshot 2026-10-20.