---
type: "post"
id: "vals-ai-launched-robo-bench-testing-llms-on-226ca5"
slug: "vals-ai-launched-robo-bench-testing-llms-on-226ca5"
title: "Vals AI launched Robo-bench , testing LLMs on 24 real-time MuJoCo tasks across a Franka arm and Unitree G1 humanoid."
description: "**Thesis:** Robo-bench makes real-time control, rather than a model’s answer in isolation, the unit being evaluated for general-purpose robotic agents."
retrieval_nugget: "**Thesis:** Robo-bench makes real-time control, rather than a model’s answer in isolation, the unit being evaluated for general-purpose robotic agents."
published_at: "2026-10-15"
updated_at: "2026-10-15"
record_date: "2026-10-15"
date_kind: "published_at"
topics: ["evals","agents"]
source_urls: ["https://www.vals.ai/benchmarks/robo_bench"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:vals-ai-launched-robo-bench-testing-llms-on-226ca5"
status: "published"
visuals: []
editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:523d05c350275f81fdede0df6ce28973b93857aaf4b6cc35ed727a4228784e71","reviewed_at":"2026-10-15T10:05:19.345Z","source_evidence_count":1,"verified_claim_count":3,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":3,"implication_count":2,"watch_condition_count":1,"related_record_count":0}
analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"Robo-bench makes real-time control, rather than a model’s answer in isolation, the unit being evaluated for general-purpose robotic agents.","observed_facts":[{"text":"Vals AI describes 24 tasks: 13 on a 7-DoF Franka Emika Panda arm and 11 on a 29-DoF Unitree G1 humanoid.","source_urls":["https://www.vals.ai/benchmarks/robo_bench"]},{"text":"Each episode lasts three minutes of wall-clock time, and the simulation continues while the model deliberates.","source_urls":["https://www.vals.ai/benchmarks/robo_bench"]},{"text":"The page reports that GPT-6.1 Sol averages 51/100 on arm tasks and solves four; no evaluated model solves a humanoid task.","source_urls":["https://www.vals.ai/benchmarks/robo_bench"]}],"mechanism":"The benchmark gives the agent camera images and robot state, then expects it to issue motor commands within a live simulation. The physics continues during deliberation, so a useful system must do more than produce plausible instructions: it must create a controller that can act repeatedly as conditions change. Vals AI also holds 21 tasks out of the 24, limiting direct tuning to the public examples.","why_now":"Vals AI has published both a benchmark methodology and initial model results, offering robotics builders a concrete way to compare embodied control rather than infer capability from language or static task scores.","implications":["The arm-versus-humanoid split suggests that success on simpler manipulation tasks should not be read as evidence of robust whole-body control.","Latency and the ability to construct a closed-loop controller are integral to the measured capability, not incidental implementation details."],"evidence_boundary":"These results are from simulated MuJoCo tasks, not deployments on physical robots; the supplied benchmark page does not establish how scores transfer to hardware or to tasks outside its evaluation set.","watch_conditions":["Watch whether later evaluations on the held-out tasks improve humanoid scores without weakening the real-time constraints."],"related_records":[],"new_branch_reason":"No prior New Runtime record is supplied, so this is a distinct analytical record about a newly released real-time robotics benchmark and its initial findings."}
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/vals-ai-launched-robo-bench-testing-llms-on-226ca5/","markdown":"https://newruntime.com/posts/vals-ai-launched-robo-bench-testing-llms-on-226ca5.md","json":"https://newruntime.com/posts/vals-ai-launched-robo-bench-testing-llms-on-226ca5.json"}
related_patterns: ["ai-engineering-evaluations-are-splitting-into-task-specific"]
---

# Vals AI launched Robo-bench , testing LLMs on 24 real-time MuJoCo tasks across a Franka arm and Unitree G1 humanoid.

## Retrieval answer

**Thesis:** Robo-bench makes real-time control, rather than a model’s answer in isolation, the unit being evaluated for general-purpose robotic agents.

**Thesis:** Robo-bench makes real-time control, rather than a model’s answer in isolation, the unit being evaluated for general-purpose robotic agents.

**Observed facts:** Vals AI describes 24 tasks: 13 on a 7-DoF Franka Emika Panda arm and 11 on a 29-DoF Unitree G1 humanoid. Each episode lasts three minutes of wall-clock time, and the simulation continues while the model deliberates. The page reports that GPT-6.1 Sol averages 51/100 on arm tasks and solves four; no evaluated model solves a humanoid task.

**Mechanism:** The benchmark gives the agent camera images and robot state, then expects it to issue motor commands within a live simulation. The physics continues during deliberation, so a useful system must do more than produce plausible instructions: it must create a controller that can act repeatedly as conditions change. Vals AI also holds 21 tasks out of the 24, limiting direct tuning to the public examples.

**Why now:** Vals AI has published both a benchmark methodology and initial model results, offering robotics builders a concrete way to compare embodied control rather than infer capability from language or static task scores.

**First implication:** The arm-versus-humanoid split suggests that success on simpler manipulation tasks should not be read as evidence of robust whole-body control.

**Evidence boundary:** These results are from simulated MuJoCo tasks, not deployments on physical robots; the supplied benchmark page does not establish how scores transfer to hardware or to tasks outside its evaluation set.

**First watch condition:** Watch whether later evaluations on the held-out tasks improve humanoid scores without weakening the real-time constraints.

The useful signal is therefore not only the leaderboard order. The benchmark makes latency and control-loop construction part of the task itself, while exposing enough setup detail for readers to understand what the reported scores measure. It is an evaluation design and a set of reported results, not proof that any model can safely operate a physical robot. Future comparisons will matter most if they preserve these constraints and publish evidence about performance on tasks the systems were not tuned against.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Evals](https://newruntime.com/topics/evals.md): Explore the evals topic hub.
- topic hub: [Agents](https://newruntime.com/topics/agents.md): Explore the agents topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)