PulseAugur
EN
LIVE 07:29:00

Codex-trained agent 'Latent' outperforms Claude Code, hits 95.5% on ARC-AGI-3

By PulseAugur Editorial · [1 sources] ·

A new AI agent, "Latent," trained using Codex, has demonstrated superior performance compared to Claude Code's own training harness. This agent achieved a 95.5% success rate on the ARC-AGI-3 benchmark, surpassing Opus 5. Additionally, Microsoft has released an open-source unit-test agent designed to analyze code repositories, and Cloudflare has launched a Chromium-free browser specifically for agent use. AI

IMPACT This development highlights advancements in agent capabilities and benchmark performance, potentially influencing future AI development and tool creation.

RANK_REASON The item discusses a new AI agent's performance on a benchmark and the release of related open-source tools. [lever_c_demoted from research: ic=1 ai=1.0]

Read on Mastodon — mastodon.social →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

Codex-trained agent 'Latent' outperforms Claude Code, hits 95.5% on ARC-AGI-3

How we ranked this

Signal score
0 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
The item discusses a new AI agent's performance on a benchmark and the release of related open-source tools. [lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
model release, product
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
62 days old
Aged out of breaking-news scoring windows; ranking reflects the durable signal from the full source set.

Full methodology in our editorial standards.

COVERAGE [1]

  1. Mastodon — mastodon.social TIER_1 English(EN) · Steenbergen_apps ·

    📰 Latent — The skill file that beat its own training harness • A Codex-trained skill file outperforms Claude Code's own • Prime Intellect's agent hits 95.5% on

    📰 Latent — The skill file that beat its own training harness • A Codex-trained skill file outperforms Claude Code's own • Prime Intellect's agent hits 95.5% on ARC-AGI-3, running Opus 5 • Microsoft open-sources a unit-test agent that reads your repo first • Cloudflare ships a bro…

    LINKS latent.steenbergen.ai/…/2026-08-07 latent.steenbergen.ai/archive

RELATED ENTITIES

RELATED TOPICS