Mistral est évidemment une brique sur laquelle nous nous appuyons. Le modèle est encore perfectible, et nous y travaillons.
GENERATIVE AI
-
GPT-5 Pro needs transparency in code execution and analysis
By
–
That said, one thing OpenAI needs to do is let me see the analysis being done by GPT-5 Pro. I can't confirm its work without being able to see the python code it ran and what the results were. The fact that the error it spotted was real helps, but the thinking trace is obscured.
-
Deep-Thinking AI Models Undersold: Market Potential Unrealized
By
–
GPT-5 Pro and Gemini 2.5 Pro Deep Think are both very impressive models for hard problems. I think they were both undersold during their respective launches, in part because I am not sure the labs themselves really understand the market for a slow, "deep-thinking" model, yet.
-

R-4B: Auto-Thinking MLLMs via Bi-Mode Annealing and RL
By
–
R-4B Incentivizing General-Purpose Auto-Thinking Capability in MLLMs via Bi-Mode Annealing and Reinforce Learning
-
ChatGPT Pro subscription value assessment and pricing models
By
–
Only 15/month https://
help.openai.com/en/articles/12
003714-chatgpt-business-models-limits
… so for me, pro is still worth it -

Jet-Nemotron and PostNAS Optimize LLM Efficiency-Accuracy Trade-offs
By
–
Jet-Nemotron & PostNAS: Redefining the Efficiency–Accuracy Frontier in LLMs @NVIDIA
’s latest research introduces Jet-Nemotron, a family of hybrid-architecture LMs powered by Post Neural Architecture Search (PostNAS). Instead of costly pretraining from scratch, PostNAS starts -
GPT-5 Pro Teams Plan Alternative Pricing Option
By
–
Btw, if you want GPT-5 Pro, but don't want to pay $200/month, you can get a Teams (called Business) plan for a bit more money (2 seats minimum) and you get some GPT-5 Pro access + you can top up the credits for more requests.
-

Grok Code Lead 60% Higher Usage Than Claude Sonnet
By
–
Grok Code lead increased to 60% higher usage than Claude Sonnet
-
Image Generation Flaws Need Comprehensive Benchmark Testing
By
–
Yes, that makes sense. Image generation is deeply flawed today. Would love to see a range of these questions used across a variety of imagegen and done over time. Would be an interesting benchmark.
-
New In-House AI Models: Token Architecture and Efficiency
By
–
Dove into our two new in-house models, token sculpture, scale vs efficiency, and more with @ReedAlbergotti and @semafor this week