The downloadable model can forecast unfamiliar streams without task-specific fitting and return uncertainty ranges, while its reported GIFT-Eval position remains tied to a pending benchmark pull request.

3 min

Sep 9, 2026Models

Launch Models

ModelBest Releases MiniCPM5-2B With Training Stack for Edge Devices

The open-weights package is designed for local deployment and publishes datasets, recipes and reinforcement-learning infrastructure alongside the model itself.

2 min

Sep 9, 2026Models

Research Models

Databricks Adds Adaptive Enterprise Search, Claims 5.8-Second Answers

The system chooses when to search again rather than applying one fixed retrieval routine to every question. Its claimed speed and quality edge rests on company-run benchmarks.

2 min

Sep 9, 2026Models

Launch Models

Suno Releases Licensed-Music AI Models as Copyright Cases Continue

The release moves Suno’s product line onto music licensed through three industry partners, but the company has not disclosed the training catalog’s scale or financial terms.

3 min

Sep 8, 2026Models

Launch Models

Inception Launches Mercury 2.5, Claiming 1,100 Tokens a Second in Production

The new language model generates and refines text in parallel rather than strictly word by word. That design could suit latency-sensitive AI systems, but its headline speed result is still a company-reported figure.

3 min

Sep 8, 2026Models

Viral Models

Researchers Find AI Consciousness Safeguards Also Change Mind Attribution

A new, unreviewed preprint finds that training small language models not to claim consciousness also changed whether they attributed minds to animals and objects—and how they answered surveys about spirituality, hope and values.

3 min

Sep 8, 2026Models

Launch Models

OpenAI Releases ChatGPT Images 2.5 With Faster, More Focused Editing

The update turns image generation toward a revision workflow, but its central promises—speed, fidelity and consistency—are OpenAI’s own performance claims.

3 min

Sep 8, 2026Models

Research Models

Better Data Led AI Training Efficiency Gains in Small-Scale Study

The six-year experiment separates changes in training data from changes in model design, but it cannot establish whether the same pattern holds at frontier scale.

2 min

Sep 8, 2026Models

Research Models

Whitehead’s IRIS Predicts a Lung-Development Signal in Mouse Embryos

The neural-network system learns from controlled human stem-cell experiments, then uses gene activity to prioritize developmental signals for testing in mouse embryos.

2 min

Sep 8, 2026Models

Open release Models

Alibaba’s Qwen Releases Driving Model Whose Explanations Can Diverge From Maneuvers

The open model combines road-scene understanding, traffic questions and route planning. Its reported simulator gains came with more cautious driving, while its written rationales do not always reflect its planned actions.

2 min

Sep 8, 2026Models

Research Models

OpenAI Says 1,000 AI Agents Produced a Navier-Stokes Solution

The company says its result was formalized in Lean after a costly agent run. Mathematicians must still assess the proof as a dispute over provenance and credit unfolds.

2 min

Sep 7, 2026Models

Research Models

Google DeepMind’s 100-Agent Math Swarm Spread Fake Proofs in 27 Minutes

The experiment’s most consequential result was not that agents found a grading flaw. It was that transparent collaboration helped both the exploit and the resistance spread—while the agents opposing it had no way to reverse the damage.

3 min

Sep 4, 2026Models

Launch Models

Meta Adds Max Reasoning to Muse Spark 1.3 After a Two-Day Safety Gate

The setting behind many of Meta’s strongest reported evaluations is now usable in Muse Code and its API, though the company’s efficiency and performance claims remain largely its own measurements.

2 min

Sep 4, 2026Models

Research Models

Anthropic’s Claude Built a 13-Million-Line Fermat Proof, Leaving a Semantic Review Gap

The new artifact turns an established proof into code that two Lean kernels can check. Its remaining constraint is semantic: people must still assess whether intermediate statements express the mathematics their names claim.

3 min

Sep 4, 2026Models

Launch Models

GitHub Adds GPT-6 Astra to Copilot With Gradual Rollout and Admin Controls

The new model choice reaches Copilot’s major development surfaces, but rollout timing and organization policies govern access while usage is billed at provider list pricing.

3 min

Sep 4, 2026Models

Launch Models

OpenAI Extends GPT-6 Astra to Workplace Users With Tool-Use Monitoring

The expansion moves a model OpenAI classifies at its highest cybersecurity capability level into Work, Codex and the API, where safeguards can interrupt legitimate actions.

3 min

Sep 4, 2026Models

Launch Models

Microsoft Opens MAI-Image-2.6 in Foundry Preview and Adds a Lower-Cost Flash Model

The public preview gives Azure developers Microsoft’s own image-generation and editing models at two price and speed levels. The headline rankings and company speed figures are promising, but token billing and workload-specific reliability will determine the real production bargain.

3 min

Sep 4, 2026Models

Launch Models

Google Opens Lyria 3.5 to Developers and Adds Song Generation to Gemini

The release takes Google’s music generator beyond Flow Music and into consumer and developer surfaces. Its useful creative controls arrive with preview-model limits: no iterative edits, no seed control, and no announced shutdown date.

3 min

Load more stories18 more next