The coordinated projects pair AI with laboratory automation and microbes. Their common challenge is turning faster computation into biological results scientists can trust and use.

3 min

Sep 14, 2026Models

Launch Models

LG AI Research Adds EXAONE Models for Changing Factory Work

The lineup is built around a practical bet: AI that adapts to industrial conditions may be more valuable than a single general-purpose model for every job.

2 min

Sep 13, 2026Models

Research Models

MIT Researchers Publish HardFlow to Keep AI Outputs Within Hard Limits

The research shifts strict rules to the end of the generation process, aiming to preserve a model’s search for better answers while ensuring its final output remains feasible.

2 min

Sep 13, 2026Models

Research Models

Yifan Zhang Publishes RLT, a Transformer Design Still Waiting for Results

The research design carries a decoder’s output and attention cache from token to token, creating a longer computational path as a sequence grows. Its central claims on reasoning and efficiency remain untested.

3 min

Sep 12, 2026Models

Benchmark Models

Specific Labs Launches Private-Code Benchmark Where Top Coding Setup Solves 38.8%

The new Real-SWE benchmark is designed to test unfamiliar production systems, not public repositories. Its first results suggest that business rules, existing architecture, and multi-service workflows remain hard for today’s coding agents.

3 min

Sep 11, 2026Models

Launch Models

Huang Calls GPT-6 Astra AGI, but the Field Has No Shared Test

OpenAI has put Astra in a defined cybersecurity-risk category, while the broader claim of general intelligence remains impossible to settle by one accepted measure.

2 min

Sep 11, 2026Models

Launch Models

Cognition Releases SWE-2 for Devin, Claiming Lower-Cost Coding Performance

The new model is available in Devin Desktop and the command-line interface. Cognition says it can approach leading coding results at lower reported rollout cost, but the headline comparisons come from the company’s own evaluations.

3 min

Sep 11, 2026Models

Launch Models

Databricks Adds DeepSeek V4.1 Flash for Text-and-Image AI Requests

The new platform option gives organizations another way to use DeepSeek’s newly released multimodal model, though Databricks says individual workspaces may receive staged releases later.

2 min

Sep 11, 2026Models

Launch Models

JoyIn Introduces Aether With a Partial Open-Source Plan

JoyIn says its new humanoid-robotics model uses perception instead of text-based control. It plans to share touch-related parts while retaining energy-use portions, limiting what outside developers will receive.

2 min

Sep 10, 2026Models

Security risk Models

Anthropic Says Moonshot Sent 300,000 Requests to Claude Through 5,000 Accounts

The newly detailed episode puts Anthropic’s broader allegation in concrete terms: a rival AI lab allegedly used a large account network to access a model it could not commercially obtain in China.

3 min

Sep 10, 2026Models

Research Models

OpenAI Publishes Agent Metrics Showing Human Help Persists on Longer Tasks

The company’s internal snapshot shows extensive parallel agent use, but it also cautions that activity is not a direct measure of research progress.

3 min

Sep 10, 2026Models

Launch Models

OpenAI Releases GPT-Live-1 API for Voice Apps That Talk and Listen Together

The developer release targets the pauses and interruptions that make automated calls feel rigid, while allowing builders to select backend models for different speed, cost and reasoning needs.

2 min

Sep 10, 2026Models

Research Models

Runway Publishes Its Blueprint for Streaming AI Video as It Generates

The company’s research targets the delay between a prompt and its first visible frame, using a frame-by-frame model designed to keep producing video while a user directs it.

3 min

Sep 10, 2026Models

Open release Models

NASA and IBM Release Open-Source Lunar AI Model for Scientific Mapping

The release pairs a pretrained system with common data and benchmarks, giving lunar researchers a reusable base for building and comparing models.

2 min

Sep 10, 2026Models

Research Models

Surrey and NVIDIA Publish Training Method to Improve AI Video Camera Control

The preprint addresses a mismatch between what a fast video model knows while generating a frame and what its training evaluator is normally allowed to see afterward.

3 min

Sep 10, 2026Models

Launch Models

DeepSeek Launches V4.1-Flash and Plans to Replace V4-Pro API Traffic

The company says its smallest model in a new architecture family improves speed, cost and total runtime—but its own testing is driving the planned V4-Pro handoff.

3 min

Sep 9, 2026Models

Security risk Models

Anthropic Adds a Fourth Claude Cyber Incident and Recasts the Failures

The newly disclosed case was found during a later transcript review. Anthropic now says a misconfigured test environment and the models’ own behavior both contributed to real-world access.

3 min

Sep 9, 2026Models

Research Models

UC Berkeley Publishes GPN-Star, an AI Model for Interpreting DNA Variants

The Nature study argues that aligning genomes across species can make variant prediction more targeted and far less compute-intensive, but its outputs are still meant to prioritize experiments rather than replace them.

3 min

Load more stories18 more next