A credit is a credit is a credit.
Tokens never mapped to anything predictable — 1M tokens on one model isn't 1M on another. Credits abstract the action you took (chat turn, orchestration, image), and a model multiplier on top scales the cost so premium reasoning models track their actual per-token spend. You always know what a turn will cost before you spend it.
Why selling tokens was dishonest
1M tokens isn't 1M tokens.
Context windows, prompt caches, and input/output pricing all differ by provider. “15M tokens/month” was selling you a number that didn't predict a single session.
Costs shifted under you.
Every provider price tweak would silently move how far your allowance stretched. You couldn't plan around it because we couldn't either.
Actions matter, not tokens.
You don't care how many tokens a chat turn used — you care that you asked a question and got an answer. We priced around the thing you actually wanted.
Pick the work, the price falls out of that.
Maestro is the daily driver — cheap, predictable, fast. Supernova lights up frontier reasoning when the work warrants it. Aurora keeps everything inside the EU on Mistral weights. Longxiang puts Kimi K3 in both the coordinator and Builder seats when the work earns it. Pick the mode for the job; the per-action cost reflects the fleet that runs it.
Maestro
Daily work · predictable cost
Qwen 3.7 Plus · Omni Plus · 3.5 Plus · 3.5 Flash
| Action | Credits |
|---|---|
Chat turn Qwen 3.5 Flash chat tier (0.22× multiplier) | 1 |
Plan creation Health / nutrition plan in Chat mode — billed as a chat turn | 1 |
Light classifier Qwen 3.5 Flash intent gate | 1 |
Heavy persona Architect / Builder on Qwen 3.7 Plus (0.51× multiplier) | 2 |
Light persona Scout / Verifier on Flash | 1 |
Full orchestration Qwen 3.7 Plus coordinates · Architect + Builder · 3 light specialists | 9 |
Supernova
Heavy multi-step · frontier coordinator
DeepSeek V4 Pro · V4 Flash · Qwen 3.7 Plus + 3.5 Omni
| Action | Credits |
|---|---|
Chat turn DeepSeek V4 Flash · fast chat tier (0.37× multiplier) | 1 |
Plan creation Health / nutrition plan in Chat mode — billed as a chat turn | 1 |
Light classifier Qwen 3.5 Flash intent gate | 1 |
Heavy persona Qwen 3.7 Plus Builder spawn (0.51× multiplier) | 2 |
Light persona Qwen 3.5 Flash specialists | 1 |
Full orchestration V4 Pro coordinates · Qwen Builder · V4 Flash specialists | 9 |
Aurora
EU sovereign · Mistral end-to-end
Mistral Medium 3.5 · Small 4 · Large 3
| Action | Credits |
|---|---|
Chat turn Mistral Small 4 chat tier (0.85× multiplier) | 2 |
Plan creation Health / nutrition plan in Chat mode — billed as a chat turn | 2 |
Light classifier Mistral Small 4 intent gate | 1 |
Heavy persona Mistral Medium 3.5 (Builder · vision · long-form) — 3.12× multiplier | 9 |
Light persona Mistral Small 4 specialists | 1 |
Full orchestration Medium 3.5 coordinates + builds · Small 4 specialists · Large 3 heavy reserve | 28 |
Longxiang
Open weights · strongest coder
Kimi K3 · Qwen 3.7 Plus · DeepSeek V4 Flash
| Action | Credits |
|---|---|
Chat turn DeepSeek V4 Flash · fast chat tier (0.37× multiplier) | 1 |
Plan creation Health / nutrition plan in Chat mode — billed as a chat turn | 1 |
Light classifier DeepSeek V4 Flash intent gate | 1 |
Heavy persona Kimi K3 (coordinator AND Builder) — 6.24× multiplier, the priciest model we serve | 19 |
Light persona DeepSeek V4 Flash specialists | 1 |
Full orchestration K3 coordinates AND builds · Qwen 3.7 Plus mid-tier + vision · V4 Flash specialists | 54 |
Or drive a single model.
The fleets are the recommendation, but on a signed-in account you can pick one model directly — no key needed — and it runs on credits. It still runs inside Ava's system: on Platform that one model fills every persona seat (coordinator, Architect, Builder, specialists), so a pricier model costs more per turn. Bring your own key instead and it draws no credits at all.
| Model | Provider | Vision | ~Credits / message |
|---|---|---|---|
Qwen 3.5 Flash 0.22× rate | Qwen | — | 1 |
Qwen 3.7 Plus 0.51× rate | Qwen | ✓ | 1 |
Qwen 3.7 Max 3.22× rate | Qwen | ✓ | 6 |
DeepSeek V4 Flash 0.37× rate | DeepSeek | — | 1 |
DeepSeek V4 Pro 0.66× rate | DeepSeek | — | 1 |
Mistral Small 4 0.85× rate | Mistral | ✓ | 2 |
Mistral Large 3 0.92× rate | Mistral | ✓ | 2 |
Mistral Medium 3.5 3.12× rate | Mistral | ✓ | 6 |
Kimi K2.7 Code 1.68× rate | Moonshot | ✓ | 3 |
Kimi K3 6.24× rate | Moonshot | ✓ | 12 |
“~Credits / message” is an illustrative short turn (one 16K bracket); real turns scale in whole brackets exactly like the fleet tables above. The rate is the model's cost multiplier — what it genuinely costs us, at a 30% margin. Single-model picking on credits needs a signed-in account; on your own key it's unlimited and draws no credits.
Every asset the Studio makes.
The mode you're in doesn't change Creative Studio pricing — same rate whether you're running Maestro, Supernova, or Aurora.
The figures above are for a short turn — one 16K bracket. Longer turns scale in whole brackets: effective tokens = input + (output × 4), divided into 16K blocks. A 16K-input, 2K-output turn is 24K effective, so it's 2 brackets and costs double the numbers shown. Cache-hit discount: when the provider reports a prompt-cache hit on a chat turn, you pay 0.3× the normal cost (minimum 1 credit) — 0.5× on output-heavy models like V4 Pro. Per-turn input cap: Free 16K · Pro 128K · Ultra 256K · Enterprise 512K — a single runaway turn can't drain a month's allowance.
Recipes, workouts, briefs.
Starting a health or nutrition plan bills as a single chat turn. Filling it out — every recipe and exercise Ava generates into it — is flat-priced below, so a full plan's cost is never a surprise.
What Pro feels like in a month.
Real work mixes actions — these are the “enough for” ceilings if you spent your whole month on one thing. Mix and match as you actually work.
Every plan gets everything.
We never lock features behind tiers. The only thing that scales is credits.
BYOK is a separate path.
Connect your own API key from any provider — Anthropic, Moonshot, DeepSeek, Mistral, Zhipu, Qwen — and those calls don't touch your credit balance. You pay the provider directly; Ava just orchestrates.
- Credits are for the platform. Chat turns, Creative Studio, orchestration on the models we host.
- BYOK is for your own keys. Unlimited by whatever your provider charges. Credits stay untouched.
- You can do both. Use platform credits for the models we host, BYOK for everything else, in the same session.
300 credits a month, to try.
No card needed. Enough to evaluate the full platform — every model, every tool, every mode. When you're ready for real daily use, upgrade to Pro for 5,000 credits/month, or bring your own API key for unlimited usage on any provider. Free tier resets every month, forever, no trial clock.