Anthropic AI R&D - LLM Training Hard Variant Average Speedup: leaderboard
Harder variant of the LLM training task in Anthropic's internal AI R&D rule-out suite, starting from already-optimized reference code with a neutral prompt; average speedup (x), unbounded, higher is better. Introduced in Table 2.3.5.A of the Claude Opus 5 System Card (July 2026); only Mythos 5 and Opus 5 were run.
Metric: Average Speedup (x). Source: www-cdn.anthropic.com. Saturation forecast: Around October 2028. 2 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 | 14.19 | #16 |
| 2 | Claude Mythos 5 | 8.36 | #8 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=anthropic-ai-r-d-llm-training-hard-variant-average-speedup · How It Works · Data refreshed daily, snapshot 2026-11-05.