News
Mistral AI Says It Found No Evidence of Unauthorized Access
Mistral AI says it investigated a claim alleging unauthorized access to its systems and found no evidence supporting it, but the brief statement leaves key…
Prompt Blueprints / Topics
The latest in AI.
Releases, research and the stories shaping AI. Reporting across the companies and models that matter.
More stories
AI releases
Ternary Bonsai 2 27B Reports 98.2% of Qwen3.8’s Score at 5.9 GB
PrismML says Ternary Bonsai 2 27B is a 5.9 GB model based on Qwen3.8 27B, with 98.2% of its reported aggregate benchmark performance under Apache 2.0.
Leaks & rumours
What an X Post Reports About Gemini 4 Pro
A single X post attributes unusually large context and output limits, persistent memory, autonomous tools, malware handling and robot control to Gemini 4 Pro.…
News
What Dream-RSI Changes in AI Exploration
Dream-RSI is described as an orchestration layer that replays an AI agent’s past discovery attempts to improve its exploration policy without changing the…
AI releases
QuiverAI Introduces Arrow 2.0 for Editable SVG Graphics
QuiverAI introduces Arrow 2.0 for generating, editing, vectorizing and animating editable SVG graphics, alongside the higher-refinement Arrow 2 Telos.
AI releases
Union Alpha is free for a week, says OpenCode
OpenCode describes Union Alpha as a stealth model for agentic coding with image support and says it will be free for one week.
News
Claude Cowork and chat are merging into one experience
Anthropic says Claude Cowork and chat will become one experience, with Pro and Max users expected to receive the update over the next few weeks.
News
What the DeepMind Institute Says It Will Study
Demis Hassabis says the DeepMind Institute will expand interdisciplinary research on AGI’s potential effects on the economy, science and society.
News
Meta One Plans and Pricing: What Each Tier Includes
Meta One adds paid features and expanded AI access across Instagram, Facebook, WhatsApp and Meta AI, with plans for individuals, creators and businesses.
News
OpenAI Reports AI-Safety Talks With Anthropic and Google
OpenAI policy chief Chris Lehane says the company has discussed AI safety with Anthropic and Google DeepMind for several weeks, while supporting proposed US…
Leaks & rumours
Gemini 3.8 Live Reportedly Appears on Google Cloud Page
A post on X reports two Gemini 3.8 entries on a Google Cloud quota and metrics page, but it does not establish a launch, customer access or model capabilities.
Leaks & rumours
Claude Opus 5.2 May Be Tested in Claude Code
An X post suggests that a model related to Claude Opus 5.2 may be undergoing greyscale testing in Claude Code under the broader “Opus 5” name, while web…
AI releases
Kimi K2.8 Preview: What Changes in Kimi Code
Kimi K2.8 Preview is fully rolled out in Kimi Code with the existing model ID, adjustable thinking effort and up to 1 million tokens of context.
Leaks & rumours
What We Know About Google’s antigravity-preview-09-2026 Model
A post on X presents antigravity-preview-09-2026 as a new Google model, while suggesting a coding focus and updates to Antigravity and Jules agents without…
News
OpenAI Shifts Safety Upstream With Pre-Run Safety Cases and Pacing
OpenAI is moving safety governance upstream into the training process, formulating pre-run safety cases for frontier reinforcement learning and calling on…
News
Satya Nadella’s Conditions for Pursuing Superintelligence
Satya Nadella says advanced AI should help humanity, remain under human control and be broadly distributed across countries, communities and companies.
News
What Dario Amodei Means by “Pacing the Frontier”
Dario Amodei proposes slowing frontier AI development enough for safety work, independent oversight and international coordination to keep pace with…
AI releases
Cognition’s SWE-2 Targets Better Coding Performance at Lower Cost
Cognition’s SWE-2 adds medium, high, and max effort levels, with company-reported gains in coding benchmarks and efficiency over SWE-1.7.