Models & Quota
Switch freely across 13 models; usage is metered by call cost.
Available models
The model selector below the chat box is grouped by brand. Ratio = relative cost (lower is cheaper); uploading images requires a vision-capable model:
| Brand | Model | Ratio | Vision | Parsing |
|---|---|---|---|---|
| Doubao | Seed 2.0 Mini | 0.25× | Yes | Chat only |
| Seed 2.1 Lite | 0.5× | Yes | Yes | |
| Seed 2.1 Pro | 2.5× | Yes | Yes | |
| DeepSeek | V4 Flash | 0.5× | No | Yes |
| V4.1 Flash | 2.5× | Yes | Yes | |
| V4 Pro | 5.5× | No | Yes | |
| Xiaomi | MiMo 2.6 Flash | 1× | Yes | Yes |
| MiMo 2.6 Pro | 3× | Yes | Yes | |
| MiniMax | M3 | 2.5× | No | Yes |
| Kimi | K2.7 Code | 4.5× | No | Chat only |
| K2.8 Preview | 8× | Yes | Yes | |
| GLM | 5.3 Flash | 0.5× | Yes | Chat only |
| 5.3 | 4.5× | No | Chat only |
Ratio = relative cost (lower is cheaper). To upload an image you must pick a vision-capable model, otherwise you'll be prompted to switch.
If the model you picked is temporarily unavailable (retired by the vendor, not in the plan, or failing), Vara finishes the chat or parse with the default model, notes which model was actually used next to the result, and meters usage by that model. Models connected with your own API key are never swapped; a failure is reported as is.
Usage quota
Metered by call cost (pricier models burn faster), capped by three sliding windows rather than calendar days, so a fresh subscription can use its full allowance the same day. Each tier's allowance relative to Free:
| Tier | Relative allowance | Notes |
|---|---|---|
| Free | 1× | Baseline |
| Pro | ~3–4× | Unlimited experiments · all models |
| Pro+ | ~15–20× | ≈ 5× of Pro |
| API (own key) | Unlimited | You pay for tokens |
The allowance is governed by three sliding windows (5 hours / 7 days / 30 days), each looser on higher tiers. The sidebar shows your 5-hour usage live; weekly / monthly allowances live under Settings → Billing. When you hit a cap you're briefly blocked; access restores automatically as the earliest usage slides out of the window, no need to wait for the 1st of the month.
On Pro and Pro+ you can add your own model key (OpenAI, Claude, Gemini, DeepSeek, Kimi, Qwen and more, or any OpenAI-compatible address) under Settings → AI → My model. Chats and parsing that use your key don't count against the allowance above; your provider bills you directly. Image understanding still uses Vara's model.