ThursdAI - The top AI news from the past week

📆 ThursdAI – Jul 31, 2025 – Qwen’s Small Models Go Big, StepFun’s Multimodal Leap, GLM-4.5’s Chart Crimes, and Runway’s Mind‑Bending Video Edits + GPT-5 soon?

70 snips
Aug 1, 2025
A whirlwind of AI advancements steals the spotlight this week! New Qwen models from Alibaba are setting benchmarks, while Runway pushes video capabilities to new heights. The excitement around GPT-5 is palpable, with discussions about emerging multimodal technologies. Innovators are racing ahead, incorporating efficient cost models and local services. Plus, breakthroughs in 3D world generation are bridging gaming and AI creativity. It's a landscape rich with developments that promises to reshape our digital experience!
Ask episode
AI Snips
Chapters
Transcript
Episode notes
INSIGHT

GLM 4.5 Benchmarking Critique

  • GLM 4.5 shows decent reasoning but benchmarks and release comparisons feel incomplete and cherry-picked.\n- Full transparency and benchmarking against all major open source leaders like Quen would build more trust in such releases.
INSIGHT

Multimodal Open Source Reasoning

  • Open source open-ended reasoning models have begun including multimodal vision capabilities in many new releases.\n- Vision benchmarks remain a challenge due to the variety and complexity compared to traditional text benchmarks.
ANECDOTE

Crush Powers Massive Quen 3 Runs

  • Nisten ran a full 480B Quen 3 model on 8 GPUs using Crush and achieved incredible throughput of around 60,000 tokens per second.\n- This setup easily handled sub agent workflows that usually take hours in minutes, highlighting the power of dedicated hardware and optimized tools.
Get the Snipd Podcast app to discover more snips from this episode
Get the app