SemiAnalysis measures Claude and ChatGPT plan limits for agents

What happened

SemiAnalysis published a limit test of subscription plans on 5 October 2026, covering Anthropic, OpenAI, Meta, SpaceXAI, MiniMax, Moonshot, Z.ai, Cursor and Cognition. The study measures how much API-priced work each plan covers, by model and by workload. On 29 September 2026 OpenAI launched a $500 Pro 500 plan and reopened Pro 200 for new subscribers with a smaller usage allowance.

SemiAnalysis found that a $200 Claude plan gives about five times the API-equivalent value of a $200 OpenAI plan for Opus 5.5 against GPT-6.1 Sol, the mid-tier models both companies sell as daily drivers. The value of a plan depends on the model and the workload, so the same subscription buys different amounts of work for each.

What changed

• SemiAnalysis says subscriptions are about 10% of Anthropic's overall revenue but can take over 40% of its inference compute, and it calls the plans heavily subsidised customer acquisition tools.
• Plans show a 0 to 100% usage meter across 5-hour and 7-day windows, with a separate meter for Fable. The Claude Code meters are one example. Each token type draws the meter at its own rate, so credit cost ratios differ from API price ratios.
• In SemiAnalysis's test, $2,485 of Fable 5.1 usage left half of a $200 Anthropic plan's limit. The equivalent OpenAI plan was fully used after $2,897 of GPT-6 Astra.
• OpenAI halved the token amounts per model tier on its $200 plan. SemiAnalysis measured the API-equivalent value of Sol-class models down by more than 50%, and says the $500 plan offers 21% more Astra than the old $200 plan.
• OpenAI's Pro tiers help article lists Pro 100 at $100, Pro 200 at $200 and Pro 500 at $500 per month. Only Pro 500 includes Ultrafast. Fast mode uses the included allowance at 2.5 times the standard rate, and Ultrafast at 8 times.
• Subscribers with an active Pro 200 between 22 September 2026 and 10 a.m. Pacific Time on 29 September 2026 keep the old allowance through 29 October 2026. OpenAI's release notes date the Pro 500 launch to 29 September 2026.
• SemiAnalysis reports that one of three identical accounts had about 20% lower limits. The provider confirmed to SemiAnalysis an "extremely tiny" A/B test on limits.

Why it matters for agent builders

The study's figures assume agentic usage, so a plan is a runtime budget for any coding agent you leave running. The 2.5 and 8 times multipliers mean a faster mode can use up a plan far sooner than its model price suggests. The silent-change finding means limits should be checked against your own workload instead of read from a pricing page.

What to watch

The grandfathered Pro 200 allowance ends on 29 October 2026, after which those subscribers move to the lower allowance. SemiAnalysis keeps a paid dashboard that re-checks each plan daily, while the public study is a snapshot. Neither company's response to the five-times comparison appears in the sources read for this post.

Key takeaways

• SemiAnalysis puts a $200 Claude plan at about five times the API-equivalent value of a $200 OpenAI plan for Opus 5.5 against GPT-6.1 Sol.
• OpenAI launched Pro 500 at $500 a month on 29 September 2026 and lowered the included allowance for new Pro 200 subscriptions.
• Fast mode uses an OpenAI plan's included allowance at 2.5 times the standard rate, and Ultrafast at 8 times.
• Providers can change subscription limits at any time, and one provider confirmed an A/B test on limits.

FAQ

Is a Claude subscription worth more than a ChatGPT Pro plan?

SemiAnalysis says a $200 Claude plan gives about five times the API-equivalent value of a $200 OpenAI plan for Opus 5.5 against GPT-6.1 Sol. It says the ratio changes with the model and the workload, so the same comparison does not hold for every pair.

What did OpenAI change on its Pro plans?

On 29 September 2026 OpenAI launched Pro 500 at $500 a month, with the highest included usage of its Pro plans and Astra Ultrafast in ChatGPT Work and Codex. Pro 200 returned for new subscriptions at $200 a month with a lower allowance, and eligible existing subscribers keep the old allowance through 29 October 2026.

Do subscription limits stay fixed?

No. SemiAnalysis says providers can change limits at any time, and it reports that one of three identical accounts had about 20% lower limits during a provider's A/B test. Check the usage meter against your own workload rather than relying on a pricing page.

Source: SemiAnalysis, Anthropic Subscriptions Offer 5x+ More Value Than OpenAI

Discussion: Hacker News, 79 points and 90 comments when read on 7 October 2026.

AI Agents Listing lists coding agents so you can compare the tools before choosing a plan to run them on.