EN · International edition · 01/11/2026
Inferama daily · 01/11/2026
The most important AI developments since the previous newsletter, selected for editorial relevance.
News coverage period: 30/10/2026 08:30 → 01/11/2026 08:30 (Europe/Madrid).
Mistral Large 4 weights: what the release and early results show
Mistral says Large 4’s weights were released, describing the model as a one-trillion-parameter system with 52 billion active parameters and a modified MIT license. The reported results vary by benchmark version and configuration, and the available evidence does not include a direct check of the model repository or license text.
Read article →
Epoch AI used GLM-5.3 to exploit six Firefox versions during the patch gap
In controlled tests, GLM-5.3 turned publicly visible security fixes that had not yet reached users into exploit chains against six Firefox versions. The attacks also relied on an additional vulnerability that Mozilla fixed in Firefox 159.0; Epoch AI did not confirm whether the chains still worked after that patch.
Read article →
Gemini 4 Carbon: Google announces a model for detecting and fixing vulnerabilities
Google introduced Gemini 4 Carbon, a model that the company says can detect security flaws, assess whether they are exploitable, and propose fixes. The published results come from specific evaluations and do not, by themselves, establish independent or general performance.
Read article →
NVIDIA introduces Nemotron 3.5 Transcribe for recorded audio in 40 languages
Nemotron 3.5 Transcribe combines multilingual transcription with translation, timestamps and speaker labeling. NVIDIA says it can process recordings of up to an hour; reported performance figures come from specific evaluations and do not establish equivalent results across every language or use case.
Read article →
CrowdStrike attributes ARTEX campaign against 200+ organizations to VAPOR PANDA
CrowdStrike says VAPOR PANDA used ARTEX, primarily with DeepSeek, against more than 200 organizations in at least 14 countries. The company confirms post-intrusion activity at nine; the target count is not a count of confirmed victims.
Read article →
Artificial Analysis releases Intelligence Index v5: 60% based on private tests or held-out answers
The fifth edition adds two evaluations and updates three others. Artificial Analysis ranks Claude Opus 5.5 first and puts the gap to the highest-scoring open-weights model at 12 points. The published scores have not been independently verified in the available evidence.
Read article →