NVIDIA AVO hits 100% on ARC-AGI-3 with no instructions across 183 levels
NVIDIA has published benchmark results for its general-purpose coding agent NVIDIA AVO on ARC-AGI-3, an interactive reasoning benchmark designed to stress adaptive problem solving under sparse guidance. According to the company post on X, AVO achieved a perfect score: it completed all 183 levels across every one of the 25 public environments without instructions, explicit rules, or stated goals. NVIDIA describes AVO as an agent that continuously inspects, plans, implements, and evaluates, using memory, tools, and execution feedback to accumulate progress across tasks. The disclosure is first-party and concise, so operators should treat the perfect score as a vendor claim until independent replication, fuller methodology, and peer scrutiny appear.
NVIDIA AVO hits 100% on ARC-AGI-3 with no instructions across 183 levels
NVIDIA says its general-purpose coding agent NVIDIA AVO scored 100% on the ARC-AGI-3 interactive reasoning benchmark. NVIDIA AVO completed all 183 levels across all 25 public environments with no instructions, explicit rules, or stated goals.
Key takeaway
NVIDIA AVO claims a flawless ARC-AGI-3 run—183 levels across 25 environments—without any provided instructions or goals, setting a new public bar for agentic coding benchmarks.
What happened
NVIDIA says its general-purpose coding agent NVIDIA AVO scored 100% on the ARC-AGI-3 interactive reasoning benchmark, completing all 183 levels across all 25 public environments.
The company states AVO figured out what to do with no instructions, explicit rules, or stated goals, and characterizes the agent as continuously inspecting, planning, implementing, and evaluating with memory, tools, and execution feedback.
Evidence
NVIDIA AVO scored 100% on ARC-AGI-3
X · attributed
NVIDIA says its general-purpose coding agent NVIDIA AVO scored 100% on the ARC-AGI-3 interactive reasoning benchmark.
AVO completed all 183 levels across 25 public environments with no instructions
X · attributed
NVIDIA AVO completed all 183 levels across all 25 public environments with no instructions, explicit rules, or stated goals.
NVIDIA describes AVO as an inspect-plan-implement-evaluate agent using memory and tools
X · attributed
NVIDIA AVO continuously inspects, plans, implements, and evaluates, using memory, tools, and execution feedback to build on what it learns along the way.
Why it matters
If the result holds under independent checks, a zero-instruction perfect ARC-AGI-3 sweep would shift how teams compare general-purpose coding agents against instruction-heavy copilots.
Limits and uncertainties
The packet contains only NVIDIA's first-party X post; no independent verification or related articles are included.
One source snippet in the packet is truncated ('Oversaturat'), so additional context from the full thread may be missing.
Practical implications
Treat the 100% ARC-AGI-3 claim as an early vendor signal rather than a settled eval standard until methodology and replication details arrive.
What to watch
Independent ARC-AGI-3 replication attempts or a fuller technical write-up from NVIDIA on AVO's benchmark setup and scoring.
Original reporting: Our general-purpose coding agent just scored 100% on the ARC-AGI-3 interactive reasoning benchmark.
NVIDIA AVO completed all 183 levels across all 25 public environments, figuring
