If the initial benchmarks scores (and graphs used for PR) showcased a 3x reduction in size for the same performance, I think the broader public reception would have been less tepid. Just looking at this, it just seems 7% behind other existing models in the 4.x series…
GENERATIVE AI
-
Kling AI Unveils Multimodal Kling O1 Model
By
–
Kling AI released the Kling O1 model with multimodal understanding that can unify input across texts, images, and videos!
— 🚨 AI News | TestingCatalog (@testingcatalog) 1 décembre 2025
AI video week 🔥 https://t.co/r7WXgVtLRf pic.twitter.com/HtXMzX5ZuhKling AI released the Kling O1 model with multimodal understanding that can unify input across texts, images, and videos! AI video week
-
Opus 4.5 Model Review: Improved Writing and Coding Capabilities
By
–
Opus 4.5: 7.5-8/10 helpful. I finally trust this model to write for me and it actually has good judgement/taste as to what matters. For coding, it feels like it can just work forever and not get stuck in the same vibe coding doom loops as previous models. Some things are
-
Runway Releases Gen-4.5 Video Generation Model
By
–
Runway released Gen-4.5, a new SOTA video gen model which scored 1,247 points in the Video leaderboard, setting.
— 🚨 AI News | TestingCatalog (@testingcatalog) 1 décembre 2025
A new top 1 👀 https://t.co/lE2nsxnSKR pic.twitter.com/KGQyyaK6SURunway released Gen-4.5, a new SOTA video gen model which scored 1,247 points in the leaderboard, setting. A new top 1
-
DeepSeek-V3.2 outperforms GPT-5 on benchmarks after release
By
–
DeepSeek-V3.2 dropped last night and absolutely crushes GPT-5 on benchmarks
-
Whisper Thunder Emerges as Google Competitor in AI Market
By
–
whisper thunder reveals itself. looks pretty good. i'm glad google has competition here, was worried they are running away too fast!!! https://t.co/QkQ0vFn3lg
— Soumith Chintala (@soumithchintala) 1 décembre 2025whisper thunder reveals itself. looks pretty good. i'm glad google has competition here, was worried they are running away too fast!!!
-
Post-training bottlenecks solved through refined methods and data
By
–
"The lesson is post-training bottlenecks are solved by refining methods and data" Zhibin Gou (@zebgou) If Gemini-3 proved continual scaling pretraining, DeepSeek-V3.2-Speciale proves scaling RL with large context. We spent a year pushing DeepSeek-V3 to its limits. The lesson is post-training bottlenecks are solved by refining methods and data, not just waiting for a better base. — https://nitter.net/zebgou/status/1995462720078934213#m
-
DeepSeek Releases DeepSeek-V3.2 and V3.2-Speciale Models
By
–
DeepSeek released DeepSeek-V3.2 and DeepSeek-V3.2-Speciale models with a top tier performance according to the benchmarks! Upgrade time
-
P-Image and P-Image Edit Models Now Available on Replicate
By
–
Run them here: P-Image: http://
replicate.com/prunaai/p-image P-Image Edit: -
P-Image-Edit: Fast Production-Ready Image Editing at One Cent
By
–
P-Image-Edit makes instant image editing production-ready Handles multi-image batch edits, precision masking, and style transfers faster than any competitor Just a cent per generation.