Every's vibe check: GPT-6 Astra impresses at writing but trails Anthropic's Fable on long-running tasks
every · x · 2026-10-03
- Every published a new Vibe Check pitting OpenAI's GPT-6 Astra against Anthropic's Fable.
- Writing & software operation: Astra's first draft, written from a single prompt, fooled Every's CEO Dan Shipper into thinking a human wrote it; it shows clear gains in writing, operating software, and visual design.
- Long-task weakness: Shipper notes Astra tends to misread prompt intent and adds too many "bells and whistles," making it less reliable than Fable for big delegation tasks.
- Head-to-head: Senior editor Jack Cheng asked both to digitize his handwritten journals into an app — the results revealed each model's strengths.
- Verdict: GPT-6 Astra is a big upgrade with some bad habits; Fable still has better instincts for building a product.
More from coding & agent
- Running Grok Bot on multiple Macs at once to scale your AI bot army — Baconbrix · 2026-10-03
- Real-world agent test: one nails church revenue analysis one-shot — MikkoH · 2026-10-03
- Agent workflow: triage Slack bugs, auto-open draft PRs when info suffices — gabrielchua · 2026-10-03
- Meta Muse hits 5M downloads as AI feedback loops could build a fast moat — RihardJarc · 2026-10-03
- Context.dev launches Highlights web-scraping API built for AI agents and LLMs — DanielLockyer · 2026-10-03
- Arize ships hosted MCP server, explains MCP vs CLI vs skills — aparnadhinak · 2026-10-03