Loading TechWhoKnows…
Reporting, analysis and primary-source coverage on Swe Bench.

Models & Research
OpenAI has launched MedChemBench, a new benchmark designed to assess how effectively AI models can interpret molecular structures, reason from experimental data, and propose new drug candidates in medicinal chemistry. GPT-6.2 Astra achieved an 84.5% score.
Clara Tan

Models & Research
Hugging Face launches harness adapters, small LoRA modules that adapt any model to a specific RL evaluation harness, cutting training cost by 80% and impr…
Clara Tan

Models & Research
RadixArk releases DaisyRose-27B, built on Qwen3.8-27B with 1,200 curated RL tasks on 144 H200 GPUs, raising Terminal-Bench 4.1 pass@1 from 5.6% to 12.0%.
Clara Tan

Models & Research
Reson8 launches Rumble-2-Flash, a 0.5B parameter multilingual speech recognition model supporting 45 languages with controllable output formatting, availa…
Clara Tan

Models & Research
Microsoft announced seven new MAI models at Ignite 2026, including MAI Code 2 and MAI Code 2 Flash for GitHub Copilot, plus MAI Cyber 2 for security workf…
Clara Tan

Models & Research
NVIDIA releases Nemotron 3.5 Super VL model, LitmusBench v0.2 dataset with 50,000 chemistry questions, and post-training recipes for scientific agents.
Clara Tan

Enterprise & Cloud
dbt Labs, Snowflake, and researchers have released Data-Eng-Bench, a comprehensive open benchmark assessing AI agents on live SQL, dbt, and DBA tasks.
Ryan Chua

Enterprise & Cloud
A rigorous architectural and latency comparison of modern secrets storage, dynamic rotation overhead, ephemeral credential issuing, and audit compliance a…
Ryan Chua

Agents & Tools
Google's developer update details how Gemini 4 Hydrogen pairs with Carbon inside Antigravity Orchestrator to execute parallel repository edits 2.5x faster…
Jin

Big Tech & Platforms
From software-locked EV battery capacity and heated seats to synthetic smartphone camera processing and benchmark tuning, how consumer tech brands monetis…
Jin

AI Safety & Policy
Anthropic just built an AI so powerful they are refusing to release it to the public. I spent the weekend digging through their massive 244-page System Ca…
Jin

Models & Research
Why developers are switching to Google's open-weight Gemma 4 models: Apache 2.0 commercial licensing, native multimodal audio on edge devices, parameter s…
Jin