Quota burn simulator
Click a query to send it to all seven models at once. Every meter starts at 0% and fills as you go, so you can see how fast each model uses up a 5-hour session and a weekly allowance.
| # | Query | Effort | Sonnet 5.5 | Opus 5.5 | Fable 5.1 | Luna 6 | Sol 6 | Sol 6.1 | Astra 6 |
|---|---|---|---|---|---|---|---|---|---|
| No queries sent yet. Each row will show the session % that query used on each model. | |||||||||
Estimates, not measurements. Token counts scale a baseline (Opus 5.5) by per-model thinking and verbosity factors anchored on published max-effort comparisons; Luna, Sol and Fable 5.1 factors are assumptions (Fable 5.1 is modeled slightly leaner than Opus 5.5, based on Firecrawl’s measured reasoning counts). Prices are list API rates per million tokens (thinking billed as output). Neither Anthropic nor OpenAI publishes plan limits in tokens, so session and weekly allowances are modeled as API-dollar budgets: $3 / $20 on the $20 tier, scaled 5x and 20x for higher tiers, applied equally to both vendors. Excludes caching, tool calls, and growing conversation context. The tokenizer approximates BPE splitting in the browser (no vendor tokenizer runs here); Claude counts are modeled at 1.15x the OpenAI-style count, so expect roughly ±15% against real API usage fields.