AI Research & Science
Peer-reviewed breakthroughs, university studies, and lab discoveries — explained in plain English. AI Maestro tracks the frontiers of machine learning research, neuroscience meets AI, and the science driving the next wave of intelligent systems.
JetBrains Releases Mellum2.1: A 12B MoE Open Model for Coding Agents
JetBrains has released Mellum2.1, an open model built for coding agents and fast sub-agents. Mellum2.1 is a 12B mixture-of-experts thinking model from…
Top stories
Claude Haiku 5.5 arrives with massive price cuts proving the AI pricing arms race is far from over
yesterday
Google bets Gemini can turn casual players into game developers with new Playground feature
yesterday
Google invests millions in Mark Zuckerberg’s efforts to create a ‘virtual cell’
yesterdayMistral Large 4
2d agoMore ai research & science
21 GPU’s benchmarked running a small TTS model (vram peak: 5GB)
Two GTX 1080 Ti’s ran the model in real-time, while a single RTX 3090…
18 May 2026
Cursor’s Composer 2.5 matches Opus 4.7 and GPT-5.5 benchmarks at a fraction of the cost
Source Read original →
18 May 2026
Benchmarked Kokoro 82M vs Supertonic 3 TTS on CPU
Wanted a real head to head on the two TTS models that actually run…
18 May 2026
The Open Agent Leaderboard
The Open Agent Leaderboard How good are general purpose AI agents? We built an…
18 May 2026
Podcast: The Physical Politics of the Internet with Britt Paris
As you scroll around the web, how often to you think about the physical…
18 May 2026
Quantizing MTP KV Cache = free lunch?
With the MTP llama.cpp implementation in the Qwen3.6/3.5 models more VRAM is required for…
18 May 2026
Qwen 3.6 27B on 24GB VRAM setup: backend comparisons, quant choice and settings (llama.cpp, ik_llama.cpp, BeeLlama, vllm)
TL;DR best setup I tested on a RTX 3090 24 GB: ik_llama.cpp + Qwen3.6-27B-MTP-IQ4_KS.gguf…
18 May 2026
Big new memory tool with local benchmarks
NOT MINE: https://github.com/rtk-ai/icm Knowledge retention: Agent recalls specific facts from a dense technical document…
18 May 2026
I built a coding agent that gets 87% on benchmarks with a 4B parameter model, here’s how
I was frustrated that every coding agent (OpenCode, Cursor, Claude Code) assumes you’re running…
18 May 2026