GitLab Transcend - Intelligent orchestration
Mini toolsBIG-Bench Hard
The 23-task subset of BIG-Bench where prior LLMs underperformed humans. Mix of logical, algorithmic, and language-understanding tasks.
Reasoning
Text
accuracy
Max 100.0%
Released Oct 2022
6
Results
6
Models scored
91.6%
Top: Seed 1.5
84.6%
Median
Best results
Top primary scores; one row per model.
Frontier over time
Each dot is one model result; the line traces the running best score.
All results
Showing all configurations including non-primary alternates. · Show only primary
| # | Model | Score | Conditions | Eval date | Source | Flags |
|---|---|---|---|---|---|---|
| 1 | Claude Sonnet 3.5 | 93.1% | 3-shot · CoT · standard | Jun 20, 2024 | self reported | |
| 2 | Seed 1.5 | 91.6% | — | Jan 22, 2025 | self reported | primary |
| 3 | Gemini 1.5 | 89.2% | 3-shot · CoT · standard | May 1, 2024 | self reported | |
| 4 | Nova Pro | 86.9% | 3-shot · CoT | Dec 3, 2024 | self reported | primary |
| 5 | Claude Opus 3 | 86.8% | 3-shot · CoT | Oct 22, 2024 | self reported | primary |
| 6 | Gemini 1.5 Flash | 85.5% | 3-shot · CoT · standard | May 1, 2024 | self reported | |
| 7 | Gemini Ultra | 83.6% | 3-shot · CoT · standard | Dec 6, 2023 | self reported | |
| 8 | Nova Lite | 82.4% | 3-shot · CoT | Dec 3, 2024 | self reported | primary |
| 9 | Nova Micro | 79.5% | 3-shot · CoT | Dec 3, 2024 | self reported | primary |
| 10 | Claude Haiku 3 | 73.7% | 3-shot · CoT · standard | Mar 4, 2024 | self reported | |
| 11 | LLaMA 2 | 51.2% | 3-shot | Jul 19, 2023 | paper | primary verified |
| 12 | LLaMA 2 70B | 51.2% | 3-shot | Jul 11, 2023 | paper |
