CALVIN ABC→D leaderboard
Long-horizon, language-conditioned benchmark that trains on environments A, B and C and tests zero-shot on unseen environment D. The official leaderboard is static, and recent papers report averages near 4.8 of 5.
Simulation · Franka Panda · PyBulletReport an error
4.78 / 5Best · MMaDA-VLA
24Models ranked
Near saturationSaturation
Nov 18, 2026Numbers checked
| # | Model | Type | Score | Date | Source |
|---|---|---|---|---|---|
| 1 | VLA | 4.78 / 5, self-reported by the authors, not a leaderboard entry | Mar 2026 | ||
| 2 | VLA | 4.75 / 5, self-reported by the authors, not a leaderboard entry | Feb 2026 | ||
| 3 | VLA | 4.73 / 5, second-hand number, not confirmed at the source | Nov 2025 | ||
| 4 | NS-VLABUPT | VLA | 4.72 / 5, self-reported by the authors, not a leaderboard entry | Mar 2026 | |
| 5 | VLA | 4.69 / 5, self-reported by the authors, not a leaderboard entry | Jul 2026 | ||
| 6 | VLA | 4.53 / 5, self-reported by the authors, not a leaderboard entry | Sep 2025 | ||
| 7 | VLA | 4.48 / 5, self-reported by the authors, not a leaderboard entry | May 2026 | ||
| 8 | VLA | 4.43 / 5, self-reported by the authors, not a leaderboard entry | Oct 2025 | ||
| 9 | VLA-AdapterOpenHelix Team | VLA | 4.42 / 5, self-reported by the authors, not a leaderboard entry | Sep 2025 | |
| 10 | UniVLA (BAAI)CASIA & BAAI | VLA | 4.41 / 5, self-reported by the authors, not a leaderboard entry | Jun 2025 | |
| 11 | World-action | 4.33 / 5, self-reported by the authors, not a leaderboard entry | Dec 2024 | ||
| 12 | Policy | 4.28 / 5, self-reported by the authors, not a leaderboard entry | Dec 2024 | ||
| 13 | VLA | 4.25 / 5, self-reported by the authors, not a leaderboard entry | Dec 2024 | ||
| 14 | VLA | 4.08 / 5, self-reported by the authors, not a leaderboard entry | Jan 2025 | ||
| 15 | Policy | 4.04 / 5, self-reported by the authors, not a leaderboard entry | Aug 2024 | ||
| 16 | VLA | 4.01 / 5, self-reported by the authors, not a leaderboard entry | Dec 2024 | ||
| 17 | UniVLA (OpenDriveLab)OpenDriveLab | VLA | 3.80 / 5, self-reported by the authors, not a leaderboard entry | May 2025 | |
| 18 | VLA | 3.66 / 5, self-reported by the authors, not a leaderboard entry | Oct 2024 | ||
| 19 | Policy | 3.61 / 5, self-reported by the authors, not a leaderboard entry | Mar 2025 | ||
| 20 | Policy | 3.35 / 5, self-reported by the authors, not a leaderboard entry | Feb 2024 | ||
| 21 | Policy | 3.06 / 5, self-reported by the authors, not a leaderboard entry | Dec 2023 | ||
| 22 | World-action | 3.00 / 5, self-reported by the authors, not a leaderboard entry | Nov 2025 | ||
| 23 | VLA | 2.68 / 5, self-reported by the authors, not a leaderboard entry | Jan 2026 | ||
| 24 | VLA | 2.48 / 5, self-reported by the authors, not a leaderboard entry | Nov 2023 |
Metric: Average completed chain length (0–5) over 1,000 five-step instruction chains, ABC→D split. One result per model: RoboFYI re-run, then official leaderboard, then paper, then third party. Runs that deviate from the official protocol are listed on the model page but not ranked.
Wrong or missing number? Report an error