ChatGPT vs Claude
How ChatGPT and Claude compare on constraint stability — usage limits, pricing, model access and transparency over time.
Constraint stability only — pricing, limits, model availability, transparency. Not a capability comparison.
01 / SIDE-BY-SIDE
| Tool | Verdict | Pricing | Limits | Models | Transparency | 90d | Last change |
|---|---|---|---|---|---|---|---|
| ChatGPT | Stable | Stable100 | Stable99 | Stable97 | Stable100 | 2 | Increases model availability for free tier users, removing previous access restrictions. (Aug 2026) |
| Claude | Watch | Stable100 | Stable91 | Stable92 | Shifting87 | 13 | Anthropic announced tightened biology-domain safeguards for Fable 5; announcement headline only, no published detail on which requests are affected. (Aug 2026) |
02 / WHICH CHANGES THE RULES LESS?
Claude's cap system has more moving parts than ChatGPT's, and it has adjusted them more quietly. It added two 7-day weekly caps atop its rolling 5-hour limits in August 2025, then in March 2026 silently made the 5-hour allowance drain faster on weekday mornings — a change confirmed only later, affecting about 7% of users at session limits — before reversing that peak throttling and doubling the 5-hour limits in May 2026. Its sharpest shock came in June 2026, when a US government export-control directive forced Fable 5 and Mythos 5 offline for all customers worldwide, with access restored by July 1. ChatGPT's history is simpler and more openly reactive: GPT-5's August 2025 launch cut Plus to 80 messages per 3 hours, backlash doubled that to 160 in September, and its other moves — ads on Free and Go, legacy models retired from the picker, both February 2026 — were visible. Both fielded near-identical 'without notice' complaints about peak-time limits. The split: ChatGPT changed less and announced it, but introduced ads and removed model choice; Claude now posts clearer limits (fixed weekly reset times) yet owns the record of silent mid-day throttling plus an externally-imposed model blackout. Day-to-day, ChatGPT's quota is easier to reason about; Claude has swung wider both ways and carries model-availability tail risk. What's being weighed is how each rations access — not how good either is once you're in.