Fable 5 Returns, Sonnet 5 Lands, Codex vs Claude — AI Daily Jun 30

687 messages · 89 active members

687
messages
89
active members
@jcartu, @geilt, @samb69
top contributors

Overview

Anthropic dominated the day with a one-two punch: Sonnet 5 launched mid-morning at ~Opus 4.7 quality for $3/mtok with promo pricing through Sept 1, then hours later Anthropic announced Fable 5 is being redeployed worldwide with export restrictions lifted. Fable 5 will count toward up to 50% of weekly usage limits through July 7 before moving to usage credits only, effectively pushing heavy users toward API pricing. Sonnet 5 drew skeptical reactions from heavy vibe coders who prefer Opus or cheaper Kimi/GLM alternatives, though it earned praise for API workloads and fast implementation after Opus-led planning. Several members hit Claude weekly limits for the first time, surfacing the tip to buy MAX directly at $200 rather than $250 via App Store. The deeper builder conversation was about coding stack architecture and workflow. Speed and steerability are beating raw intelligence — iterating fast with a mid-tier model often wins over waiting hours for a one-shot. Codex CLI 0.140.0's new /import command migrates entire Claude Code setups (skills, hooks, MCP, CLAUDE.md) to AGENTS.md. @thewildzeno detailed a multi-agent review pipeline burning millions of tokens across ~10 specialized agents with adversarial cross-model passes, while @geilt shared a docs-first architecture with 10K+ domain-organized docs that agents grep-discover on demand. @tounano reframed prompt engineering as 'prompt copywriting' — triggering attention via scoped todo-tool tasks with per-step success criteria. The GLM-5.2 + Oh My Pi + Codex stack keeps hardening as the local-first coding consensus. On creative and tooling: @mb29266 stress-tested Grok Imagine 1.5 via kie.ai and found it ~10x cheaper than Seedance, generating 15-sec clips in ~5 seconds at pennies per clip. Voice input momentum keeps building — Wispr Flow, Handy with Parakeet/Whisper, and the new Wispr Pedal for push-to-talk dictation. Indie game dev emerged as a legit side quest, with weekend MVPs in Godot and MECCHA CHAMELEON cited as proof that indie + Twitch distribution is a real wedge.

Topics

Anthropic confirmed Fable 5 is coming back with US export restrictions lifted worldwide. Included for up to 50% of weekly usage limits through July 7, then moves to usage credits only — effectively pushing heavy users toward API pricing. Builders are already planning where to spend Fable tokens: security audits, planning, complex refactors, and math-heavy analysis.

Sonnet 5 landed at roughly Opus 4.7 quality for $3/mtok with promo pricing through Sept 1, and Anthropic removed the sonnet-specific weekly limit. Heavy vibe coders question its purpose vs Opus or cheaper Kimi/GLM, while others find it strong for API workloads, planning, and fast implementation after Opus-led design. Several members simultaneously hit Claude weekly limits — buy MAX directly at $200 vs $250 via App Store to skip Apple's 30% cut.

Builders converged on speed and iteration velocity mattering more than one-shot intelligence for most coding work, with many moving from Claude Code to Codex for token efficiency. Codex CLI 0.140.0's /import migrates Claude Code configs (skills, hooks, MCP, CLAUDE.md) to AGENTS.md. Opus remains the pixel-perfect one-shot king but too slow for iteration; Grok Heavy is a fast integrator, not a strategist. The GLM-5.2 + Oh My Pi + Codex stack keeps hardening as the local-first consensus.

@thewildzeno detailed a Claude Code workflow orchestrating ~10 specialized review agents across Codex and Sonnet with adversarial cross-model passes, replacing tools like CodeRabbit and Greptile. @geilt shared a docs-first architecture with 10K+ domain-organized docs in a separate repo, grep-discoverable by agents. @tounano reframed prompt engineering as attention-triggering via scoped todo-tool sequences with explicit per-step success criteria.

@mb29266 tested Grok Imagine 1.5 via kie.ai — ~10x cheaper than Seedance, 15-sec clips in ~5 seconds at pennies per clip. Voice input momentum grew with Wispr Flow, Handy (Parakeet on PC, Whisper on Mac), and the new Wispr Pedal for push-to-talk. Indie game dev emerged as a legit side quest: @iammikevineyard shipped a Godot MVP in a weekend, and MECCHA CHAMELEON was cited as proof that indie + Twitch distribution is a real wedge.

Key Takeaways

  • Fable 5 returns worldwide with 50% weekly limit inclusion through July 7, then usage credits only — plan your high-value tasks (security audits, refactors, math) now.
  • Sonnet 5 sits at ~Opus 4.7 for $3/mtok with promo pricing through Sept 1; best for API workloads and implementation after Opus planning, less compelling vs Kimi/GLM for cost-sensitive vibe coders.
  • Buy Claude MAX directly at $200 instead of $250 via App Store to avoid Apple's 30% cut; Codex CLI 0.140.0's /import migrates entire Claude Code setups to AGENTS.md.
  • Speed and steerability are beating raw intelligence — iterating fast with a mid-tier model often wins over waiting hours for a one-shot from a smarter one.
  • Docs-first architecture (domain-organized, grep-discoverable) plus scoped todo-tool prompts with per-step success criteria dramatically improve multi-agent reliability.

Hot Threads

@samb69started

Fable 5 redeployment details and strategy for the 50% limit / API-only future

20 replies12 participants
@archimortystarted

Sonnet 5 vs Opus, Kimi, GLM — initial impressions and where it fits

15 replies8 participants
@thewildzenostarted

Multi-agent Claude Code review workflow burning millions of tokens

12 replies4 participants

Linked Items