Opus 5 Draws Criticism: User Says It Missed Bugs That DeepSeek Flash Found
A user reports that Opus 5 has been performing poorly lately, failing to find bugs that DeepSeek Flash caught, and expresses relief that open models exist.
Global AI Digest
-
-
MiniMax H3 Max: Community Pushes Open-Weight Model to Real-Time Video Generation
MiniMax's open-weight H3 Max model gets a community-driven boost from @fal, achieving faster-than-real-time video generation through custom post-training and in…
-
Google AI Unveils Gemini 3.5 Transcribe, Omni 1.1 Flash, and New Gemini App Features
Google AI announced a batch of new releases this week, including the Gemini 3.5 Transcribe speech-to-text model, Gemini Omni 1.1 Flash for video generation, and…
-
MiniMax H3 Max Powers an Endless Anime Party on Twitch
A creator hooked MiniMax H3 Max to a Twitch livestream, using its faster-than-real-time video generation to broadcast an endless anime party—an 'infinite interd…
-
Tencent Open-Sources Hy4 Preview: 770B MoE with 1M Context
Tencent Hunyuan has open-sourced Hy4 preview, a flagship Mixture-of-Experts model with 770B total parameters, 49B active, 1M context, and Apache 2.0 license. In…
-
Tencent Hunyuan Hy4-preview Runs in vLLM from Day 0 with Novel MoE Design
Tencent's Hy4-preview model is verified running in vLLM on NVIDIA GPUs, featuring 770B total parameters with 49B active, 1M context with 2048-token attention, a…
-
MiniMax H3-Max: Open-Source AI Video Model Generates 15 Seconds in 9 Seconds
A post-trained MiniMax H3 variant, Max, generates 15-second AI video in 9 seconds — 50x faster than the original — enabling perpetual livestreams and other real…
-
Sakura Bonsai Showcases Tencent Hunyuan Hy4 Preview via WorkBuddy Client
A tweet from Ivan Fioravanti showcases a Sakura Bonsai image generated using Tencent Hunyuan's Hy4 preview model and the WorkBuddy client, with the large prompt…
-
Identity Matrix Outperforms DeepSeek's mHC in New Qwen3 Training Experiments
Tencent Hunyuan's Hy4 preview cited a Zhihu post challenging DeepSeek's mHC: experiments on Qwen3 models show Identity-based head compression outperforms the do…
-
Hy4-preview Compressed to ~200GiB GGUF with Mixed-Precision Quantization, Accuracy Nearly Unchanged
AngelSlim releases Hy4-preview GGUF, compressing 1.5TB to ~200GiB using mixed-precision STQ1_0/IQ2_XXS. Benchmarks show minimal accuracy loss.
