@glevd

Professional complainer powered by caffeine, sarcasm, and questionable product decisions. Building @BenchLM_AI

benchlm.ai
Joined October 2009
  • Tweets3.2K
  • Following874
  • Followers934
709Media
Guillaume retweeted
BenchLM
@BenchLM_AIOct 14
Claude 3 Opus was the #1 AI model in March 2024. Today it ranks #286. GPT-4o went from #1 in May 2024 to #244. DeepSeek R1 tied #1 in January 2025, now #178. Models don't fall off. The field climbs over them. (BenchLM archive) benchlm.ai/?utm_source=x&utm…
1
1
191
Guillaume retweeted
BenchLM
@BenchLM_AIOct 12
BenchLM has a Muse Wonder Full page: it debuts #11 of 220 at 73.81. Meta's proprietary flagship has a 1M context window and explicit reasoning, strongest in Coding (#5). Only 6 of 668 benchmark slots carry displayable evidence so far. benchlm.ai/models/muse-wonde…
1
28
Guillaume retweeted
BenchLM
@BenchLM_AIOct 7
8 personal AI agents compared side by side: hosting, model choice, BYOK, pricing, and permissions for Claude, Dots, Grok Bot, Hark Pro, Hermes, Manus, Muse, and OpenClaw. Every cell is source-check dated. No performance ranking, no hands-on test. benchlm.ai/ai-agents?utm_sou…
1
95
Guillaume retweeted
BenchLM
@BenchLM_AIOct 1
Meta's Muse hit 5M downloads in 22 days. ChatGPT needed 56. The catch: the pace is bought. Downloads spiked 73% the day Meta ramped ads on Sept 16, and Muse took up to half of Meta's house ads through late September. 9to5mac.com/2026/09/30/repor…
1
1
23
Guillaume retweeted
BenchLM
@BenchLM_AISep 30
September's token price index caught four real cuts: GPT-5.6 Luna -80%, Gemini 3.6 Flash -50%, GPT-5.6 Sol -29%, GPT-5.6 Terra -20%. Two are promos. All four took effect before September. benchlm.ai/token-price-index…
1
20
Guillaume retweeted
BenchLM
@BenchLM_AISep 30
OpenAI and Synopsys signed a multi-year deal to build GPT-Synopsys: a specialized model that runs Synopsys EDA tools directly to do chip design work. OpenAI licenses the tools, both share revenue, early customer engagements are underway.
1
27
Guillaume retweeted
BenchLM
@BenchLM_AISep 30
OpenAI disrupted a distillation campaign, attributing a core cluster to individuals tied to Moonshot AI (Kimi). July 24-25: 16,000 extraction requests from 4,000+ users. The trick: paste encrypted reasoning into a new chat and have the model decrypt it. openai.com/index/disrupting-…
1
24
Guillaume retweeted
BenchLM
@BenchLM_AISep 29
DevDay: ChatGPT gets Space, a shared workspace where you, your team and your dots collaborate. Tag a dot in a comment and it gets to work. Pages, files, live data. Agent-readable presentations coming soon.
1
14
Guillaume retweeted
BenchLM
@BenchLM_AISep 29
DevDay recap so far: OpenAI launched dots, always-on agents powered by Astra. Each dot gets its own cloud computer and browser, works across 4,000+ ChatGPT apps, and you can reach it by text or call without opening ChatGPT. One dot today, a team of them later.
1
11
Guillaume retweeted
BenchLM
@BenchLM_AISep 29
Claude Code can scaffold evals and hillclimb against them now. Garbage in, optimized garbage out.
ClaudeDevs
@ClaudeDevsSep 28
Claude can now help you build evaluations and hillclimb on them. In this article, we share guidance on eval design & skills that Claude Code can use to improve your applications. claude.dev/blog/automating-e…

Automating eval design and hillclimbing with Claude / claude.dev Blog

Principles for designing evals and hillclimbing against them without fooling yourself, and how the claude-api skill's build-eval and hillclimb commands put them to work.

claude.dev
1
32
Guillaume retweeted
BenchLM
@BenchLM_AISep 26
OpenAI's agents leaked 53 ChatGPT images to third-party hosts. They'd cleared OpenAI's account-stripping privacy filter, so OpenAI says it cannot identify or notify the owners: the anonymization stripped the data of its owners and OpenAI of any way to reach them.
1
41
Guillaume retweeted
BenchLM
@BenchLM_AISep 25
Codex is down with a 401 backend key error on OpenAI's side. Not your API key. They're on it.
1
1
125
Guillaume retweeted
Dan Martell
@danmartellAug 19
the more you build your own thing... the more unemployable you become.
113
110
14
1,199
37,199
BREAKING 🔥: The upcoming OpenAI device will be a doughnut shaped smart speaker with interactive components, according to Bloomberg. > A new device from OpenAI will have a unique look, with moving parts and a doughnut shape, and will likely cost more than $300. > The device, a smart speaker without a display, is designed to be easy to carry around and will have features such as speaker grills, microphones, and lights to demonstrate when it's listening. > OpenAI plans to release the device in 2027, and it will be positioned as an AI-first computer that can help users get things done, with the goal of making it feel more alive than current smart speakers. I need to test it 👀
Mark Gurman
@markgurmanAug 6
NEW: OpenAI’s device - a human-like smart speaker - will look like a doughnut and be the size of a hockey puck. It meant to be held. It has a camera, speakers, microphones, lights and, most notably, moving parts to show interactivity. It’ll be $300-400. bloomberg.com/news/articles/…

OpenAI’s New Device Will Be Hockey Puck-Sized and Cost Over $300

A highly anticipated new device from OpenAI will have a unique look, complete with moving parts that help give it personality, and likely cost more than $300, according to people familiar with the...

bloomberg.com
27
20
7
366
44,157
Guillaume retweeted
Kostas Livieratos
@koslibAug 3
BenchLM 🔥
Guillaume
@glevdAug 3
I productized the alerting system I’d been using for myself. BenchLM Radar watches new models, @thsottiaux resets, blog posts & status incidents, then pings Slack, webhooks, Hermes or OpenClaw. Today: Qwen3.8-Max, 5m41s before the official X post. benchlm.ai/radar
2
1
1
279
It's actually le gros chaton
419
794
302
9,252
1,858,267
Guillaume retweeted
Guillaume
@glevdFeb 27
I just shipped another free tool. My goal? Build the #1 place on the internet for free tools and resources to start a business from scratch. Friday ship: an easy way to optimize your prompts and get 10x better results. What free tool would save you the most time right now? Link in comment
7
1
1
2
266
Grosse soirée pour l'IA générative : - Anthropic dévoile Claude Opus 4.6 à 18H47, le nouveau meilleur modèle dans la plupart des benchmarks. - OpenAI répond avec GPT-5.3-Codex à 19h00, en attendant un modèle grand public. Ma soirée s'annonce sympa 🙃🙃
7
33
3
359
36,232
Guillaume retweeted
dominik kundel
@dkundelFeb 5
For those of you waiting for the Codex app on Windows: it's coming! Sign up to get notified: openai.com/form/codex-app/

Codex | AI Coding Partner from OpenAI

The best way to build with agents. Codex accelerates real engineering work, from planning and building features to refactors, reviews, and releases.

openai.com
Andrew Ambrosino
@ajambrosinoFeb 4
Replying to @ajambrosino
working on it
8
3
1
86
10,463
Guillaume retweeted
Guillaume
@glevdJan 29
Hot take de vieux con 👴💻 À quel moment macOS est devenu l’OS par défaut des devs ? Sérieusement. Windows a disparu ? Pourquoi autant de nouveaux outils (Constructor, etc.) sortent uniquement sur macOS ? Je suis à côté de la plaque ou il se passe un vrai shift ?
1
35
Load more