Shanghai AI Laboratory InternRobotics
InternVLA-A1.5
2.69B Non-commercial 8/8 verified
Facts
- Released
- 2026-07 1 Source
- Parameters
- 2.69 B 1 Source
- Architecture
- VLM + action expert 0 Source
- License
- CC BY-NC-SA 4.0 (code repository and model card agree) 2 Source
- Commercial
- Non-commercial 0 Source
- Weights
- huggingface.co ↗ 1 Source
- Code
- github.com ↗ 1 Source
- Paper
- arxiv.org ↗ 0 Source
- Input modalities
- rgb · language · state 0 Source
- Action space
- Joints / end-effector (action expert) 0 Source
- Cross-embodiment
- InternData-A1 + LeRobot v2.1 data; RoboTwin 2.0 / LIBERO / SimplerEnv / DOMINO 1 Source
- Pretraining data (hours)
- Unverified 0 Source
- Target forms
- Single arm · Dual arm 0 Source
Architecture · Qwen3.5-2B native VLM + lightweight unified action expert; four weight sets: base / RoboTwin / LIBERO / DOMINO
Sources · 2 sources
license,arch_notes,cross_embodiment,code_url https://github.com/InternRobotics/InternVLA-A1 Official page · Fetched 2026-09-05 · No snapshot stored yet README：CC BY-NC-SA 4.0；Qwen3.5-2B backbone + lightweight unified action expert；InternData-A1；RoboTwin 2.0 / LIBERO / SimplerEnv / DOMINO
params_b,weights_url,license,release_date https://huggingface.co/api/models/InternRobotics/InternVLA-A1.5-base Model hub page · Fetched 2026-09-05 · No snapshot stored yet safetensors 2,685,264,547；license: cc-by-nc-sa-4.0；createdAt 2026-07-04；base Qwen/Qwen3.5-2B
Activity signals
Public counters fetched daily from GitHub, Hugging Face and ModelScope. Numbers only; no judgement. · fetched 2026-10-31
GitHub stars
572
+9 / 30 d
Commits · 90 d
0
last push 48 d ago
HF downloads · 30 d
152
all-time 650 · 10 likes
Weights last modified
2026-07-16
108 d ago
GitHubInternRobotics/InternVLA-A1 · 45 forks · 19 open issues · no GitHub release
ModelScopeInternRobotics/InternVLA-A1.5-base · 115 downloads · updated 2026-07-16 · same org as Hugging Face
Hugging FaceInternRobotics/InternVLA-A1.5-base
stars · last 30 d
Public benchmark scores
As printed in the source; protocols differ between papers. Full table with notes on the benchmarks page.
| Benchmark | Subset | Score | Source |
|---|---|---|---|
| LIBERO | avg | 98.9 | paper · Table 5, row 'InternVLA-A1.5 (Ours)', column 'Average' |
| LIBERO | goal | 98.6 | paper · Table 5, row 'InternVLA-A1.5 (Ours)', column 'Goal' |
| LIBERO | long | 98.4 | paper · Table 5, row 'InternVLA-A1.5 (Ours)', column 'Long' |
| LIBERO | object | 99.8 | paper · Table 5, row 'InternVLA-A1.5 (Ours)', column 'Object' |
| LIBERO | spatial | 98.6 | paper · Table 5, row 'InternVLA-A1.5 (Ours)', column 'Spatial' |
| Other benchmarks | test_success_rate | 35.2 | paper · Table 7 'Results on EBench', row 'InternVLA-A1.5 (Ours)', column Test / SR |
| Other benchmarks | total | 84.8 | paper · Table 6 'Results on LIBERO-Plus', row 'InternVLA-A1.5 (Ours)', column Total |
| Other benchmarks | zero_shot_success_rate | 27.7 | paper · Table 4 'Comparison on DOMINO', row 'InternVLA-A1.5 (Zero-shot)', column SR (%) |
| Real-robot evaluation (self-reported) | mof_long_horizon_success | 76.4 | paper · Sec. 5 real-world results text: 'On the long-horizon MOF task, InternVLA-A1.5 reaches 76.4% while pi0.5 succeeds in only 29.3% of the trials and Motus fails to complete the procedure in any trial' |
| RoboTwin 2.0 | avg | 93.2 | paper · Table 3, row 'InternVLA-A1.5 (Ours)', column 'Avg.' |
| RoboTwin 2.0 | clean | 93.3 | paper · Table 3, row 'InternVLA-A1.5 (Ours)', column 'Clean' |
| RoboTwin 2.0 | randomized | 93 | paper · Table 3, row 'InternVLA-A1.5 (Ours)', column 'Rand.' |
| SimplerEnv · WidowX (Bridge) | avg | 80.8 | paper · Table 2 'Results on SimplerEnv (WidowX). All numbers are success rates (%)', row 'InternVLA-A1.5 (Ours)', column Average |
Deployment recipe
No recipe yet. We publish only steps that actually ran on a rented GPU with a measured latency; this model has not been through that yet.
Record history
2026-09-05
Sources fetched
license,arch_notes,cross_embodiment,code_url · params_b,weights_url,license,release_date
Latency & cost
cost = p50 ms ÷ 1000 × rental ÷ 3600 × 1000
| Hardware | p50 ms | p95 ms | Peak VRAM | Cost per 1k inferences ¥ | samples |
|---|---|---|---|---|---|
NVIDIA RTX 4090 ¥1.68/h · GongJi Compute (2026) | — | — | — | — no latency data | No data |
NVIDIA RTX 5090 ¥2.4/h · Runyun 2.4 / GongJi 2.5 / Paratera Cloud 2.98 (2026) | — | — | — | — no latency data | No data |
NVIDIA A100 80GB no rental price | — | — | — | — no latency data | No data |
NVIDIA Jetson AGX Thor no rental price | — | — | — | — no latency data | No data |
NVIDIA H100 80GB no rental price | — | — | — | — no latency data | No data |
NVIDIA A800 80GB no rental price | — | — | — | — no latency data | No data |
No latency measurements for this model yet · Data comes from literature, maintainer runs, or crowd CLI uploads. None exist so far.
Compatibility
Colored only where we hold evidence. An amber dot means the vendor declares this form as a target; that is a claim, not a verified compatibility.
| Unitree G1 | Unitree H1 | AgiBot Genie G1 | AgileX Piper arm | SO-101 | Franka FR3 / Panda | Fourier GR-2 | WidowX 250 S | Google Robot（Everyday Robots） | ALOHA 2 / Cobot Magic (dual-arm) | Galaxea R1 Pro | Franka Emika Panda | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| InternVLA-A1.5 |
Legend Officially supportedCommunity reproducedFine-tunable in theoryUnsupportedUnknown Vendor-declared target form 4 / 12 cells with a definite status
Reproduction records
Did this model run on your robot? Report it with a link (log, video or repo). Records are reviewed by the maintainers; a verified “works” or “works after fine-tuning” upgrades the matrix cell to “community verified” with your link as evidence. Anonymous is fine; nothing but the fields below is stored.
No verified record yet for this model.
Clear direction, grounded judgement.