Loose posts

GPT-5.5 ran the same 1,000 tasks in both Claude Code and Codex and its scores shifted with the harness; the benchmark asks agents to rebuild arXiv figures as editable PowerPoint slides, with perception still the main bottleneck.

Against AutoResearch on 12 AI research tasks, it reaches 10.3% more reward with 50.6% fewer research attempts under the same budget.

A demo video says the modeling, rigging, texturing and animation were all done by AI, producing an animated windmill in Blender in 15 minutes.

With power limits threatening to leave AI hardware idle, Nvidia is pursuing smaller facilities, software that gets more out of existing GPUs, and deals that steer electricity to its chip customers.

Higgsfield used 5.5 million tokens to have Claude Opus 5.5 analyze an exoskeleton, find its weak points and redesign it, producing a complete 3D model in 39 minutes in one shot.

Opus 5.5 on medium effort scores 52.5% on CursorBench versus 51.8% for Fable 5.1 on max effort, at $2.91 per task against $17.28.

OpenAI warns that recursive self-improvement could accelerate beyond our collective ability to understand progress, assess risks and maintain meaningful human oversight.

Both models passed nearly all of HAProxy's tests, but Opus 5.5 finished in 9.5 hours at 51% lower cost, and the author calls it his daily driver.

Qwen-Image-2.1 ranks first among open models with 1228 points and 17th overall, 4 points behind nano-banana-pro.

The YC-backed startup automates accounting, sales and demand planning for brands including Olipop, Mars and Unilever, and has raised a $55 million Series B.

Harvard researchers found most grad school applicants now let AI write their essays, but those who did were admitted less often, in a paper titled "AI-written admissions essays are widespread but penalized."

Aimed at agents pulling stale context, the AML benchmark runs 150 software tasks each with relevant history and again with noisy history to test memory retrieval.