Robot Intelligence Index

Updated November 17, 2026102 models23 benchmarks

A unified ranking of robot policies built from 23 sim and real-world benchmarks.

  • Sim and real · 36
  • Any two benchmarks · 64
  • All models · 102
  • All
  • VLA
  • WAM
  • LLM
  • Other
  • All
  • Open
  • Closed
36 models

Overall

Best-ranked model from each lab

Rating vs speed

Median time per task attempt in this view · hollow dots are estimates

Faster, higher rated1000125015001750200020 s50 s1 min 40 s3 min 20 s8 min 20 s16 min 40 s33 min 20 sTime per task (faster ←)Rating (better ↑)Claude Fable 5.2SLA-PreviewInternW0-ΔGalaxeaVLA G0π0-FAST

Rating vs cost

Median compute cost per task attempt · hollow dots are estimates

Cheaper, higher rated10001250150017502000$0.010$0.100$1.00$10.00Cost per task (cheaper ←)Rating (better ↑)Claude Fable 5.2SLA-PreviewInternW0-ΔGalaxeaVLA G0π0-FAST
ModelRank rangeModel weights
1Claude Fable 5.2Anthropic · LLM1989± 41–2$7.60est.3 min 18 sest.4ClosedNov 09, 20262SLA-PreviewHKUST(GZ) & OLA · VLA1807± 101–8$0.041est.45 sest.2OpenNov 05, 20263Dream-1Dream Labs · WAM1756± 512–22$0.116est.2 min 20 sest.2ClosedOct 19, 20264GPT-6 AstraOpenAI · LLM1733± 252–10$17.74est.3 min 01 sest.10ClosedSep 03, 20265GPT-6.1 SolOpenAI · LLM1726± 122–11$2.15est.4 min 02 sest.5ClosedSep 29, 20266InternW0-ΔShanghai AI Lab · WAM1723± 113–10$0.028est.30 sest.3OpenSep 25, 20267MWA-PhoenixAnyverse & OpenPhoenix · WAM1690± 103–9$0.041est.45 sest.2OpenOct 21, 20268Gemini 4 ArgonGoogle · LLM1685± 143–12$3.08est.25 min 54 sest.6ClosedSep 30, 20269Cosmos3-Nano-PolicyNVIDIA · WAM1668± 94–12$0.045est.49 sest.3OpenMay 31, 202610OpenWAM-αOpenWAM team · WAM1666± 174–13$0.030est.33 sest.4OpenSep 07, 202611Claude Opus 5.5Anthropic · LLM1630± 215–13$6.17est.3 min 06 sest.7ClosedSep 22, 202612FLUX 3 ActionBlack Forest Labs · WAM1618± 354–16$0.052est.57 sest.2OpenSep 23, 202613DreamZeroNVIDIA · WAM1596± 302–17$0.048est.52 sest.3OpenFeb 17, 202614π0.5Physical Intelligence · VLA1517± 2210–16$0.030est.33 sest.18OpenApr 22, 202515Gemini 3.8 FlashGoogle · LLM1489± 212–20$3.9643 min 21 s2ClosedSep 02, 202616MolmoAct2Ai2 · VLA1411± 2612–25$0.038est.41 sest.10OpenMay 05, 202617InternVLA-A1Shanghai AI Lab · VLA1405± 3213–28$0.045est.49 sest.2OpenJan 05, 202618GalaxeaVLA G0.5Galaxea AI · VLA1396± 2712–26$0.036est.40 sest.5OpenJun 16, 202619Muse Wonder 1.0Meta · LLM1369± 3113–28$1.52est.4 min 33 sest.2ClosedOct 12, 202620LingBot-VLA v2Robbyant (Ant Group) · VLA1368± 614–26$0.036est.40 sest.3OpenJul 07, 202621GalaxeaVLA G0Galaxea AI · VLA1367± 1315–27$0.028est.30 sest.3OpenAug 30, 202522π0-FASTPhysical Intelligence · VLA1352± 1914–26$0.014est.16 sest.6OpenJan 16, 202523GR00T N1.6NVIDIA · VLA1350± 2613–28$0.077est.84 sest.3OpenSep 29, 202524Xiaomi-Robotics-0Xiaomi · VLA1347± 1617–27$0.031est.34 sest.4OpenFeb 12, 202625DeepSeek V4.1 FlashDeepSeek · LLM1346± 1715–28$0.277est.7 min 05 sest.3OpenSep 10, 202626GR00T N1.7NVIDIA · VLA1332± 2916–28$0.035est.38 sest.12OpenMar 16, 202627GLM 5.3 FlashZhipu · LLM1332± 1813–29$0.31552 min 22 s2OpenAug 26, 202628X-VLATsinghua AIR et al. · VLA1288± 3716–32$0.027est.30 sest.8OpenOct 11, 202529π0Physical Intelligence · VLA1243± 2323–33$0.029est.31 sest.15OpenOct 31, 202430StarVLAStarVLA team · VLA1231± 1416–33$0.030est.33 sest.3OpenOct 15, 202531Diffusion PolicyColumbia / TRI · Baseline1210± 2420–35$0.028est.31 sest.4OpenMar 07, 202332DM0Dexmal · VLA1206± 3011–35$0.061est.67 sest.3OpenFeb 10, 202633Spirit v1.5Spirit AI · VLA1117± 427–35$0.045est.49 sest.2OpenJan 12, 202634ACTStanford · Baseline1108± 1327–35$0.030est.33 sest.5OpenApr 23, 202335SmolVLAHugging Face · VLA1102± 1425–36$0.027est.30 sest.5OpenJun 02, 202536PaliGemma-binningPhysical Intelligence · VLA1060± 028–36$0.051est.55 sest.2OpenJun 22, 2025

Ranked as in the Robot Intelligence Index: rank and its plausible range within the chosen coverage. The ± under a rating is its spread, how much the model’s finishes differ across benchmarks, not uncertainty in the rating. Cost and time are per task attempt, median across the view’s benchmarks; est. marks Playground estimates. Methodology