Agno gemini-3.7-flash learning from gpt-5.6's notes scored 96.42 on the ARC-AGI-3 public set, beating the human baseline of 95.4.
Public source
Publisher name
Public post
In January I proposed that memory is the wrong abstraction and what agents really need is the ability to learn. Learn about users, their environment, their purpose, thei…
Company
Agno
Agno is the programming language for agentic software.
- Location
- New York, US
- Company size
- 11–50 employees
About Agno
Agno is a lightweight framework for building multi-modal Agents Github: https://git.new/agno Docs: https://docs.agno.com
See moreLatest activity
Latest activity from Agno
8 signals
Products & Services
Agno launched the new Atomic Mail toolkit, which gives an Agno agent a dedicated inbox to read and send mail, allowing it to handle email as its own participant rather than through a shared account.
Products & Services
Agno shipped OpenRouteServiceTools in version 2.8.7 to replace the estimate with a real routing engine.
Products & Services
Agno launched version 3.0 of its platform, hitting 100% on the ARC-AGI-3 public set with a score of 100.00 for an agent running GPT-5.6.
Discover more
Similar signals
Similar public activity from other companies.
Research & Knowledge
FuturePath AI
FuturePath AI reports that Astra achieved 62.7% score with ARC's standard harness and 99.9% score with OpenAI's provider adapter on ARC-AGI-3.
Research & Knowledge
The GEO Community
The GEO Community published a technical analysis examining the Prime Agent system's four-level hierarchy for model, context, computational, and durable state, persistent Python execution, recursive subagents, harness-level self-improvement through versioned prompts, memories, skills, and subagent specifications, and results across ARC-AGI-3, long-context tasks, autonomous research, and GPU-kernel coding.
Research & Knowledge
Contio
Contio achieved 99.9% performance on ARC-AGI-3 using OpenAI’s Provider Adapter.
Research & Knowledge
Microsoft
Microsoft authors report 30% to 95.5% on ARC-AGI-3 using Prime Agent and Continual Harness.
Research & Knowledge
Freelancing.