AI Briefing
TLDR AI news
The latest AI and developer news from TLDR AI, with short summaries.
Visit website →Google Launches Gemini 3.7 Flash with Enhanced Coding and Agent Performance and 50% Price Discount
Google has launched Gemini 3.7 Flash, optimized for coding and agent workflows, offering a 50% API price discount until the end of 2026.
GPT-5.6 Sol Ultrafast Mode Preview: Up to 14x Faster Speed
OpenAI has unveiled the 'Ultrafast' mode, which boosts GPT-5.6 Sol's speed by up to 14x using Cerebras technology.
How a 3-Person Team Ships Hundreds of PRs Every Week
A 3-person team reveals their workflow for shipping hundreds of high-quality PRs using agentic engineering.
As AI Safety Concerns Grow, Three Pioneers Advocate for the Need for 'Open'
Geoffrey Hinton, Fei-Fei Li, and Andrew Ng argued for the necessity of open models for innovation despite AI safety concerns.
What kinds of mathematics are LLMs good at?
Analyzes how LLMs solve mathematical challenges and whether they have a particular strength in finding counterexamples.
Hiring Agents Is the Easy Part
The core of AI agents lies not in simple hiring, but in building a management system that continuously verifies and improves work quality.
MAI-Image-2.6 Achieves 2nd Place on Arena Leaderboard
MAI-Image-2.6 secured 2nd place on the Arena text-to-image leaderboard, surpassing models from Google, Meta, and xAI.
Microsoft Releases MAI-Thinking-1
Microsoft has released MAI-Thinking-1, a high-performance reasoning model optimized for enterprise tasks.
Specula: Extending Formal Specifications for Autonomous Model Verification of System Code
Introduces Specula, a technique that extends formal specifications for system code to enable autonomous model verification.
Claude in Chrome Side Panel Evolves into Full-Fledged Collaboration Sessions
Claude has introduced full-fledged collaboration sessions that perform tasks by directly manipulating web pages through the Chrome side panel.
Manus Acquisition Falls Through
Manus has announced the resumption of independent operations after Meta's acquisition attempt failed, urging users to back up their data.
rd-signal-2: State-of-the-Art Classification Models at Production Scale
Raindrop has launched rd-signal-2 and Signal Builder, which build task-specific binary classifiers based on production traces.
How AI Watermarks Hide in Plain Text
Anthropic's text watermark is hidden through statistical biases in word choice rather than at the byte level.
Import from Other Agents
Provides a documentation index for models, agents, tools, and various features of the OpenAI API.
Model Picker is a dead end
Lovable proposes a 'control plane' approach that automatically assigns the optimal model based on task characteristics, rather than letting users select models manually.
Google's New AI Chief Inherits the Task of Chasing OpenAI and Anthropic
With the leadership change at Google DeepMind, a full-scale competition to catch up with OpenAI and Anthropic begins.
Extracting Reasoning Traces from Commercial LLM APIs
Researchers successfully reconstructed encrypted reasoning traces from commercial LLM APIs to recover hidden reasoning processes.
Nemotron 3.5 Lightning (12-minute read)
NVIDIA released Nemotron 3.5 Lightning, a 30B MoE model with 3B active parameters for low-latency AI agent tasks.
MAI-Code-1.1-Flash: Better, Faster, and a Quarter of the Cost
Microsoft AI introduced MAI-Code-1.1-Flash to GitHub Copilot, improving performance and speed while reducing costs to a quarter.
Addressable Memory for Video World Models: WorldTrace
WorldTrace is a framework that helps video world models accurately remember and utilize past visual information even when the generation length exceeds the training scope.