Claude ÷ GPT on the $20 plan, as an index
Every bar is Claude Opus 5.5 relative to GPT-6.1 Sol (GPT = 100). The first two bars are why the last two look the way they do.
GPT-6.1 Sol
Claude Opus 5.5
GPT baseline = 100
2.9× more tokens ÷ 3.1× per task = 0.93× the work · 5.6× API value ÷ 8.3× cost per task = 0.68× per dollar
Sources: SemiAnalysis Tokenomics Model, 5 Oct 2026 (tokens per plan, API-equivalent value) · Artificial Analysis Intelligence Index v4.3.2, 6 Oct 2026 (output tokens and cost per task, max effort).
The numbers behind the bars
Numbers
| Per $20/month plan | GPT-6.1 Sol | Claude Opus 5.5 | Claude ÷ GPT |
|---|---|---|---|
| Tokens included | 1.0B / month | 2.9B / month | 2.9× |
| Output tokens per AA task | 38.1k | 119.2k | 3.1× |
| Cost per task (API rates) | $0.72 | $5.98 | 8.3× |
| API-equivalent value in plan | $211 | $1,178 | 5.6× |
| ≈ Tasks per plan (token-normalized) | 100 | 93 | 0.93× |
| ≈ Tasks per plan (API-priced) | 100 | 68 | 0.68× |
Method & caveats
How to read this
- What is measured. Tokens per plan: SemiAnalysis limit-tested each plan (agentic coding until the weekly cap; mix 0.4% input / 96.6% cached / 2.6% cache writes / 0.3% output). Tokens and cost per task: Artificial Analysis ran both models on identical evaluations. "Tasks" divides the first by the second, holding the billed-token mix constant.
- The honest caveat. This compares equal effort tiers (max vs max) — the configs in the original chart. At equal intelligence instead (Opus 5.5 Medium ≈ Sol Max), Opus is the thriftier model — 25.7k vs 38.1k tokens per task — and Claude delivers 3–4× more work per plan. The token chart is incomplete, not wrong.
- Sources. SemiAnalysis — “Anthropic Subscriptions Offer 5x+ More Value Than OpenAI” (5 Oct 2026) · Artificial Analysis — GPT-6.1 Sol (Max) vs Claude Opus 5.5 (Max, Default Fallback). Deeper write-up: plan-token-reality.plan.abstergo.fyi.