Ben's Bites: Opus 5 Matches Fable 5 at Half Price, Visual Agents in tldraw
Ben's Bites · rss · 2026-07-28
This week's Ben's Bites covers major AI dynamics and tool practices:
Models & Products
- Claude Opus 5: Matches Fable 5 performance at half the price. Anthropic removed over 80% of Claude Code's system prompt with no measurable coding-eval loss.
- ChatGPT Voice: Can now run the desktop app and delegate tasks via voice, though still falls short of text for serious work.
- Kimi K3: Weights and technical report are out, now available at 50% off on Droid.
Open Source & Drama
- Following rumors of a US ban on Chinese open-weight models, NVIDIA’s Jensen Huang made his first X post backing open-source AI. While 150+ companies signed the statement, Anthropic refused and later published a defensive essay, facing community backlash for hypocrisy.
Dev Tools
- The author shares experiments with tldraw offline: using a visual canvas for agents to build widgets (weather timelines, timezone checkers) proves more effective than pure text.
- Mentions new tools like FLUX 3 unified model, ChatGPT Health integration, and the open-source OpenWorker Mac agent.
More from Models
- Kimi K3 reportedly runs on 80 RTX 5090s with 2.56 TB of VRAM — sandyyevans · 2026-07-28
- Users report Opus 5 starts sounding like Trump in base model mode — Sauers_ · 2026-07-28
- Kimi is described with a linear-attention variant and attention residuals — burny_tech · 2026-07-28
- Reddit users say OpenAI’s Sol may have degraded after release — unhappinessNvrCame · 2026-07-28
- GPT-5.6 sol is described as a full coding package for medium-to-large codebases — haider1 · 2026-07-28
- Moonshot AI reportedly seeks advanced Nvidia chips for Kimi K4 training — Polymarket · 2026-07-28