Live AI developer news, ranked and linked to original sources.
> ▌

Ben Davis

AI Revolution

DIY Smart Code

DIY Smart Code

Theo - t3․gg

Eric Michaud

AI LABS

DIY Smart Code

Discover AI

Bijan Bowen

The PrimeTime

DIY Smart Code

Github Awesome

OpenAI

Income stream surfers

Two Minute Papers

AICodeKing

Prompt Engineering
OpenAI's GPT-6.2 Astra and Sol models have debuted on Macroscope's code review benchmark, which tests defect detection across real-world pull requests. Astra claimed the #2 overall spot with an 83.2 F1 score at high effort—finishing behind only Claude Opus 5.5 max effort—while Sol took #4 and achieved the leaderboard's highest precision at 91.6%.
Odysseys2 is a web agent benchmark developed by Carnegie Mellon University and Browser Use to evaluate AI agents across 100 complex, live-web tasks lasting up to 60 minutes. An LLM judge grades final delivered files alongside screenshots and action histories to penalize reward hacking, leaving leading models completing fewer than half the tasks.
A stealth AI model codenamed Gestalt has surged to the top of Contra Labs' Frontier Design Benchmark with an unprecedented 1,931 Elo rating. Evaluated blindly by professional designers on production UI briefs, the unannounced model outperformed Claude Fable 5.2 by 218 points and cleared the human baseline.
Google DeepMind's Gemini 4 Neon claimed the top spot on Parallel's Search Intelligence leaderboard, outperforming all rival frontier models while operating at up to ten times lower cost. The mid-tier reasoning model establishes a new Pareto frontier for autonomous web research and agentic retrieval.
Google's Gemini 4 Neon went head-to-head against GPT-6.2 and Opus 5.5 across eight single-file HTML, CSS, and JavaScript site builds, completing the entire benchmark suite for $4.52. The test demonstrates that frontier models can reliably generate functional, styled front-end prototypes in single prompts at negligible API cost.
Google's mid-tier reasoning model reached the second spot on Adam Holter's HearingBench leaderboard, trailing only Gemini 4 Carbon. The benchmark challenges vision-language models to identify sounds, words, and full sentences purely from visual spectrogram images without audio input.
Google's Gemini 4 Neon has overtaken Anthropic's Claude Fable 5.2 on the Artificial Analysis Coding Agent Index, scoring 59 at extra-high reasoning effort. The mid-tier model delivers top-tier agentic coding benchmark performance at roughly half the operational cost and higher token throughput than rival frontier models.

Wes Roth

AI Search