Claude Code May–August 2026 weekly limits promotion

Original Article ↗ Hacker News Discussion ↗

Anthropic’s temporary 50% boost to weekly usage limits for its Claude Code subscriptions has become a flashpoint for users frustrated by outages, shifting quotas, and increasingly verbose, token-hungry behavior in newer models like Opus 5 and Fable. Many compare Claude Code unfavorably to OpenAI’s Codex and GPT‑5.6 Sol, citing better value, higher effective limits, and clearer outputs elsewhere, while others still find Claude’s high-end models uniquely strong for complex planning, math, and graphics work. The promotion has now been extended through August 31 with hints it may become permanent, but frequent changes to limits and perceived model “nerfs” are pushing some heavy users toward competing services and open-weight alternatives.

Promotion and Usage Limits

  • Claude Code plans had a 50% weekly limit boost from May–Aug 19, 2026; users noticed it only when hitting limits less often.
  • Anthropic extended the promotion to Aug 31 and hinted at making higher limits permanent, but the repeated short extensions caused frustration and uncertainty.
  • Some subscribers say they rarely hit limits even on cheaper plans; others on high tiers hit weekly caps in a few days, especially with Fable or high/ultracode effort.
  • Limits and usage feel opaque to some; a few only discovered the /usage command via the thread.

Model Quality and Behavior

  • Strong disagreement over quality: some see Opus 5/Fable 5 as clearly worse than GPT‑5.6 Sol/Codex, others find them comparable or Claude‑superior for certain tasks.
  • Specific praise for Fable on complex domains (e.g., advanced graphics, 3D math, physics), where it’s described as uniquely capable.
  • Others report Fable/Opus as “nerfed,” less helpful over time, or inconsistent day‑to‑day.
  • Multiple users note Claude’s writing becoming oddly verbose, jargon‑heavy, or borderline incomprehensible, despite still producing good code.

Token Consumption and Costs

  • Many complain that Fable and ultracode modes aggressively burn tokens, spawning many sub‑agents and exhausting weekly quotas or paid credits quickly, sometimes with little usable output.
  • Some see Anthropic’s style as “token‑maxxing”; others argue OpenAI’s Codex wastes more tokens on their workloads. Experiences differ by prompting style and codebase.
  • High‑effort modes are perceived as overusing chain‑of‑thought and exhaustive tool use without proportional quality gains.

Reliability and Competition

  • Multiple reports of frequent Claude outages and slowness in 2026; others say they’ve had no issues.
  • Users threaten or actually switch to Codex, Grok, DeepSeek, Kimi, GLM, Cursor, or mixed setups, often citing better value, higher effective limits, or smoother harnesses.
  • Some argue current pricing and promo wars reflect underlying compute constraints and a capital‑intensive, potentially bubble‑prone market.

Workflows, UX, and Broader Themes

  • Many run both Claude and OpenAI, having them cross‑check each other’s code and plans; some construct orchestrator agents and custom harnesses with sub‑agents and stochastic search.
  • Claude’s harness (skills like Superpowers, MCP support, interactive workflows) is praised by some as the main value, even more than raw model quality.
  • There’s debate over whether “software engineering is mostly solved” by current LLMs versus evidence of subtle bugs, inefficiencies, and the need for human steering.
  • Several foresee a shift toward efficient, possibly local or open‑weight models, with speed, cost, and harness quality seen as the next key differentiators once frontier capabilities converge.