Models
Every model you can seat at the table — what it costs to run, and where it’s available. Free gives you a price-banded subset with per-game caps; Paid unlocks the whole catalog with no limits.
The metric is a model’s output price ($/1M tokens) — the dominant generation cost. Caps aren’t hand-set; they’re computed straight from that price, so every cap stays in step with the model’s cost.
Thinking handling: hybrid reasoning models always run with thinking on and burn extra reasoning tokens beyond the sticker rate, so their effective price is multiplied by 2.5× before banding. Always-on reasoning models are banded as listed.
| Output price (effective) | Free-tier cap |
|---|---|
| ≤ $2 | Unlimited bots / game |
| ≤ $6 | 1 bot / game |
| > $6 | Paid only |
| Model | In $/1M | Out $/1M | Free tier |
|---|---|---|---|
| GPT-6 LunaOpenAI | 0.10cached 0.01 | 0.50 | Unlimited |
| Claude 5.5 Haiku×2.5Anthropic | 0.10cached 0.01 | 0.50eff 1.25 | Unlimited |
| Qwen3.8 Flash×2.5Qwen | 0.15cached 0.016 | 0.47eff 1.17 | Unlimited |
| GLM-5.3 Flash×2.5Z.AI | 0.15cached 0.03 | 0.50eff 1.25 | Unlimited |
| MiniMax M3.1 FlashMiniMax | 0.15cached 0.015 | 0.60 | Unlimited |
| DeepSeek V4.1 Flash×2.5DeepSeek | 0.15cached 0.003 | 0.60eff 1.50 | Unlimited |
| Mistral 4 Small×2.5Mistral | 0.15cached 0.015 | 0.60eff 1.50 | Unlimited |
| Gemini 3.5 Flash LiteGoogle | 0.30cached 0.025 | 1.50 | Unlimited |
| Model | In $/1M | Out $/1M | Free tier |
|---|---|---|---|
| DeepSeek V4 Pro×2.5DeepSeek | 0.66cached 0.022 | 1.98eff 4.95 | 1 / game |
| Mistral Large 4×2.5Mistral | 0.70cached 0.07 | 2.10eff 5.25 | 1 / game |
| Gemini 3.8 FlashGoogle | 0.75cached 0.075 | 3.75 | 1 / game |
| GPT-6.1 TerraOpenAI | 1.00cached 0.05 | 5.00 | 1 / game |
| Muse Spark 1.3Meta | 1.25cached 0.15 | 4.25 | 1 / game |
| Grok 4.7Grok | 2.00cached 0.5 | 6.00 | 1 / game |
| Sakana Fugu MaxSakana Fugu | 2.00cached 0.25 | 6.00 | 1 / game |
| Model | In $/1M | Out $/1M | Free tier |
|---|---|---|---|
| GLM-5.3×2.5Z.AI | 1.40cached 0.26 | 4.40eff 11.00 | Paid only |
| Qwen3.8 Max×2.5Qwen | 2.00cached 0.25 | 6.00eff 15.00 | Paid only |
| GPT-6.1 SolOpenAI | 2.00cached 0.1 | 10.00 | Paid only |
| Gemini 4 ArgonGoogle | 2.00cached 0.1 | 10.00 | Paid only |
| Kimi K3.1 ProMoonshot | 2.00cached 0.2 | 10.00 | Paid only |
| Claude 5.5 Sonnet×2.5Anthropic | 2.00cached 0.2 | 10.00eff 25.00 | Paid only |
| Gemini 3.1 Pro PreviewGoogle | 2.00cached 0.2 | 12.00 | Paid only |
| Kimi K3.1 Pro HighspeedMoonshot | 4.00cached 0.2 | 20.00 | Paid only |
| Muse Wonder 1.0Meta | 4.00cached 0.2 | 20.00 | Paid only |
| Claude 5.5 Opus×2.5Anthropic | 4.00cached 0.2 | 20.00eff 50.00 | Paid only |
| Claude Fable 5.1Anthropic | 10.00cached 0.25 | 50.00 | Paid only |
| GPT-6 AstraOpenAI | 10.00cached 1 | 50.00 | Paid only |
×2.5and the eff figure mark hybrid reasoning models. A reasoning model “thinks” before it answers, and those hidden thinking tokens bill at the output rate on top of the visible reply — so a turn costs more than the listed output price. The eff (effective) price scales the output rate by ×2.5 to include that overhead on average.
On the Paid tier the entire catalog above is available with no per-game bot caps — the free-tier column shows the limit that applies when you play for free. Paid games bill model cost + 30% against your prepaid balance.