AI daily2026-07-26

Anthropic's Opus 5 Shatters ARC-AGI Record, Tencent Open-Sources Embodied AI Models

Anthropic's Claude Opus 5 achieves 30.2% on ARC-AGI-3, nearly quadrupling the previous record, while Tencent open-sources three embodied foundation models with a novel 'three-layer brain' architecture. DeepSeek's funding round is paused after a leaked memo.

Selected
6
Sources
16
Featured
6
More
0

AI DAILY BRIEFING

Anthropic released Claude Opus 5 at half the price of its flagship Fable 5, yet it outperforms Fable 5 on multiple benchmarks including coding, agentic search, computer use, and enterprise workflows. Notably, Opus 5 scored 30.2% on ARC-AGI 3, nearly four times the previous best, demonstrating unprecedented fluid intelligence. The benchmark's developers noted that the model independently formulated reflection equations, a behavior not seen before, indicating stronger logical reasoning.

Anthropic's pricing strategy with Opus 5 undercuts its own flagship while delivering superior performance, potentially disrupting the AI model pricing landscape.

The ARC-AGI-3 score of 30.2% represents a significant leap in fluid intelligence benchmarks, suggesting that large language models are approaching new frontiers in reasoning.

Tencent's open-source release of three embodied foundation models and the 'three-layer brain' architecture could accelerate robotics research by addressing latency issues through decoupled cognition, perception, and execution.

Watch next: DeepSeek's funding pause and Liang Wenfeng's open-source stance may signal a shift in AI startup funding dynamics, while the open-sourcing of tactile data and models by NeoteAI and Fudan University could advance embodied AI's sense of touch.

Featured

SSignal87
Claude (X)
15 sources

Anthropic Releases Claude Opus 5 AI Model

Anthropic has launched Claude Opus 5, a thoughtful and proactive AI model that approaches the frontier intelligence of Fable 5 at half the price. It achieves state-of-the-art results on coding and knowledge work evaluations, and scores three times higher than the next best model on ARC-AGI-3. Opus 5 is available today on all paid plans and the Claude API.

SynthePulse InsightRead deep analysis
APriority79
AI前线
2 sources

Tencent Open-Sources Three Embodied Foundation Models; Chief Scientist Zhang Zhengyou Explains How 'Three-Layer Brain' Solves Robot Latency

At WAIC 2026, Tencent open-sourced three embodied foundation models and introduced a 'three-layer brain' architecture that separates cognition, perception, and motor control to run at different frequencies, addressing robot response delays. Chief Scientist Zhang Zhengyou emphasized that the physical world operates on multiple time scales, making this layered design a long-term stable solution.

SynthePulse InsightRead deep analysis
SSignal89
机器之心
2 sources

NeoteAI and Fudan University Release Tactile Data Technical Reports to Enhance Embodied AI's Sense of Touch

NeoteAI, in collaboration with Fudan University, has released three technical reports aiming to address the lack of tactile feedback in embodied AI by providing 30,000 hours of tactile interaction data. The team introduced a unified tactile representation model, NeoForce, and two technical approaches: -VTLA predicts future tactile evolution to improve operation success rates, while -TWAM integrates touch into world models, allowing robots to simulate tactile sensations before execution. The project's data and models will be open-sourced, potentially advancing fine manipulation in embodied AI.

SynthePulse InsightRead deep analysis
BStandard59
THE DECODER
1 sources

Anthropic's Opus 5 blows past Fable 5 and GPT-5.6 Sol on the benchmark designed to measure real intelligence

Anthropic's Claude Opus 5 scored 30.2% on the ARC-AGI-3 benchmark, nearly quadrupling the previous record of 7.8% set by GPT-5.6 Sol. The benchmark's developers noted that the model independently formulated reflection equations, a behavior not seen before, indicating stronger logical reasoning.

SynthePulse InsightRead deep analysis
APriority71
机器之心
1 sources

Valkor, Zhejiang University, and UCL Open Source Loom: A Structured Engineering State Layer for AI Coding Agents

Valkor, together with the Intelligent Computing and Software Research Center of Zhejiang University and the Software Engineering team at University College London (UCL), has open-sourced Loom. Loom addresses the 'local amnesia' problem in AI coding agents by providing a structured engineering state layer that breaks down complex delivery processes into resumable state chains, helping agents maintain context after test failures or session interruptions, thereby improving reliability in real-world engineering tasks.

SynthePulse InsightRead deep analysis
BStandard57
InfoQ
1 sources

Liang Wenfeng Halts Signing, DeepSeek's Billions in New Funding May Be Fully Suspended

DeepSeek founder Liang Wenfeng, angered by a leaked investor meeting memo, has halted the signing of a new funding round, pausing the company's accelerated $71 billion valuation fundraising and IPO preparations. The leak has prompted internal reflection, with Liang reiterating an open-source strategy and asserting that open source does not affect the business model.

SynthePulse InsightRead deep analysis
All reports