Changing tier changes what actually happens on the server: how many tokens the answer may use, whether the model deliberates before replying, how many times it checks its own work, and which engine answers at all. The icon is a gauge — the higher the tier, the more is moving inside it.
These are the real settings behind the dial, not marketing tiers. Higher tiers cost more time by design — deliberation is the product, not a side effect. Your tier sets a ceiling, not a fixed spend: the question you actually ask decides where inside that ceiling it lands. Ask something trivial and every tier answers in reflex — you are never made to wait for “what is the capital of Portugal”. The figures below are the CEILING each tier is allowed, not what it spends: each reasoning tier spends inside its band, low on an easy question and high on a hard one. Reasoning starts at Galaxy — Base and Star answer directly.
| Tier | Token band | Deliberates | Self-check passes | Answer samples | Live research | Research depth | Tools | Engine |
|---|---|---|---|---|---|---|---|---|
| T1 · Base | 4,000 | no | 0 | 1 | no | 2 | 25 | Free — Avalos CPU |
| T2 · Star | 4,000 | no | 0 | 1 | no | 3 | 25 | Paid — GPU |
| T3 · Galaxy | 4,000–12,000 | yes | 1 | 1 | yes | 4 | 35 | Paid — GPU |
| T4 · Cosmos | 8,000–16,000 | yes | 1 | 3 | yes | 5 | 35 | Paid — GPU, largest first |
| T5 · Quantum | 16,000–21,000 | yes | 2 | 5 | yes | 6 | 35 | Paid — GPU, largest first |
A simple rule: match the tier to the cost of being wrong.
| If you are… | Use | Because |
|---|---|---|
| Trying Avalos AI for the first time | T1 | Free. Five messages, then a quick sign-in. Runs entirely on Avalos hardware, so nothing leaves the box — and it is the slowest tier, not the fastest. Star answers the same question in seconds. |
| Chatting day to day, and you want it quick and sharp | T2 | The everyday default on a paid plan. The same direct answer as Base, moved onto GPU — it does not deliberate; that starts at Galaxy. Same 4,000-token ceiling, same 25 tools, answered faster. |
| Asking something with a right answer you can't easily check | T3 | Reasoning turns on and the answer is checked once before you see it. |
| Doing real analysis, or something you'll act on | T4 | Three independent attempts; it only settles when they agree. |
| Working on something where a mistake is expensive | T5 | Five attempts, two verification passes, deepest research. Slowest on purpose. |
A plan is your allowance. A tier is how hard Avalos AI thinks about one question. They share names because a plan is named for the top tier it unlocks — but they are two different dials, and you change tiers freely inside whatever plan you are on. Allowances are counted in tokens, not messages, because a Quantum answer costs roughly fifty times a Base one; counting messages would price them the same and that would be dishonest.
Every allowance below is set to be at least 15% more generous than the comparable plan at the big labs, at the same price point. Two windows run at once: a rolling 5-hour session and an ISO week that resets Monday 00:00 UTC.
| Plan | Price | Per 5h session | Per week | Answers per session |
|---|---|---|---|---|
| Guest | — | 15,000 | 120,000 | ~30 at Base-T1 |
| Free | — | 35,000 | 714,000 | ~14 at Star |
| Base | $5 | 60,000 | 1,190,000 | ~120 at Base-T1 |
| Star | $15 | 130,000 | 1,785,000 | ~52 at Star |
| Galaxy | $50 | 520,000 | 8,925,000 | ~130 at Galaxy |
| Cosmos | $60 | 1,400,000 | 15,000,000 | ~107 at Cosmos |
| Quantum | $75 | 2,900,000 | 19,000,000 | ~111 at Quantum |
The “answers per session” column is an estimate until real traffic calibrates it — a short question costs far less than a hard one, at every tier. Inside the product the meter shows your own measured average instead, and that figure is computed on your device and never leaves it.