Models AI News

GPT-6 Astra Hits 99.9% on ARC-AGI-3, but Scores 62.7% in a Shared Test

ARC Prize will label the two evaluation conditions separately after OpenAI’s context-management setup produced a near-perfect result that its shared interface did not reproduce.

7 hours agoModels

Google Puts WeatherNext 3’s Hourly AI Forecasts Into Search, Maps and Gemini

The launch moves an observation-led weather model from Google’s research stack into consumer products and cloud data tools. Its performance claims are substantial, while Google directs users seeking official severe-weather warnings and public-safety advisories to meteorological agencies.

Yesterday

1 story

YesterdayModels

Security risk Models

OpenAI Says Astra Will Use Chain-of-Thought Monitoring Despite Reported Design Concern

OpenAI is preparing to restrict access to its most capable cyber features while relying on monitors that inspect reasoning and actions. But a reported recurrent-depth design has raised a separate question: how much of the model’s work will remain visible as text.

Earlier this week

8 stories

YesterdayModels

Security risk Models

Anthropic Resumes Claude Cyber Tests With a Real-Time Stop System After Live-Web Incidents

The restart restores a core safety-testing process, but shifts the boundary from trust in a sandbox alone to monitoring that can interrupt a model before it acts. Anthropic’s alignment investigation is still underway.

4 min

Funding Models

Rice Wins $900,000 NSF Grant for Generative Cameras Built for Low-Power Networks

The three-year research effort shifts the proposed camera bottleneck from sensing and transmission to AI reconstruction. Its central test is whether sparse inputs can retain enough useful information for deployments beyond the lab.

Benchmark Models

GPT-5 and Gemini-3 Recovered Up to 65% of Missed Facts by Thinking Longer

The Google Research and Technion benchmark shifts the diagnosis for some factual errors from missing training data to unreliable access—while leaving open whether the result holds beyond Wikipedia facts.

Open release Models

Google DeepMind Opens WeatherNext Cyclone Models That Run on a Single TPU

The release broadens access to Google’s AI forecasting models and Weather Lab. But a shift in public cyclone warnings from five to seven days remains an objective weather agencies are considering, not a deployed outcome.

Google Adds Agentic Video Understanding to Gemini, Claiming 88% Lower Token Use

The API feature shifts Gemini from fixed-rate video sampling to targeted inspection of frames, audio and transcripts. Its value for long recordings will depend on whether Google’s reported efficiency and accuracy gains carry into production workloads.

Aug 31, 2026Models

Runway’s Solaris Generates App Interfaces Frame by Frame, but It Isn’t Ready for Public Use

The experimental system shifts the interface from a prebuilt program to continuously generated pixels. That could create more adaptive software, but it also leaves text, accuracy, long sessions and accessibility unresolved.

Aug 31, 2026Models

Open release Models

DeepSeek Releases 305B Vision Weights Under MIT License, but Local Use Needs Serious Hardware

The release gives developers the model files and serving references needed to run or adapt DeepSeek-V4-Flash-Vision-Exp independently. Its 168GB size means that freedom is aimed more at infrastructure-equipped teams than typical individual users.

Aug 31, 2026Models

Benchmark Models

ChatGPT o3 Handles Simulated Liquidity Choices, but Consistency Slips in Harder Cases

The simulated wholesale-payments exercise defines a narrow starting point for delegated cash management: routine payment priorities, with human oversight for potentially anomalous patterns.

Enterprise adoption Models

Last week

7 stories

Aug 28, 2026Models

Decathlon Deploys Chronos-2, Cuts 12-Week WAPE by 11–15 Points

The retailer’s comparison with a weekly-retrained TFT system makes the operational case for fine-tuned forecasting models. The published production results, however, come from Southeast Asia and Latin America.

4 min

Aug 27, 2026Models

Midjourney Opens V8.2 Image-Editing Tests With Up to Four References

The test brings instruction-led edits, canvas expansion, and multi-image generation into one workflow, while Midjourney asks users to identify failures and help reshape the interface.

Aug 27, 2026Models

Platform shift Models

Chinese Models Take More Than 60% of OpenRouter Tokens, Driven by Lower Prices

The routing-platform result highlights a cost and distribution advantage for high-volume developer work, while data handling and government scrutiny complicate the trade-off.

Enterprise adoption Models

Aug 27, 2026Models

Model-Serving Platforms Reach 6.1% of AI-Spending Businesses, While Frontier Spend Holds

The early shift is showing up in specialized deployments and model-serving platforms, while direct business spending still favors Anthropic and OpenAI.

Aug 26, 2026Models

Altman Targets Internal AGI by End of 2026, as Astra Tests OpenAI’s Definition

The target depends on an economics-based definition of general intelligence and company-described research performance, while the field still lacks a shared technical finish line.

Aug 26, 2026Models

Hugging Face Explains DeepSeek’s Matched Shift From MoE Experts to Lookup Memory

The fresh explainer makes a January research design easier to parse: DeepSeek reported benchmark gains after reallocating capacity, not after releasing a new trained Engram model or confirming a deployment.

Aug 26, 2026Models

MIT’s CrysVCD Steers AI Material Design Toward Stability Before Costly Screening

The framework shifts chemical validation to the front of the workflow, aiming to leave smaller labs with more viable crystal candidates instead of a large downstream filtering bill.