Repository navigation
AI GM Bot API Pricing
Nick Gasper edited this page Aug 26, 2026
·
1 revision
Here's a practical breakdown of what each model will cost you per session, ordered from cheapest to most expensive with performance estimates for GM-ing.
| Model | Provider | Input $/M | Output $/M | Est. GM Quality | Best For |
|---|---|---|---|---|---|
| Local (Ollama) | Self-hosted | $0 | $0 | ★★☆☆☆ (varies by model) | Zero-cost, privacy-first |
| Gemini 2.5 Flash-Lite | $0.10 | $0.40 | ★★☆☆☆ | Budget, casual games | |
| Ministral 3 3B | Mistral | $0.10 | $0.10 | ★★☆☆☆ | Edge/offline deployments |
| Mistral Small 4 | Mistral | $0.15 | $0.60 | ★★★☆☆ | Price-performance sweet spot |
| Ministral 3 8B | Mistral | $0.15 | $0.15 | ★★★☆☆ | Output-heavy sessions |
| GPT-5.6 Luna | OpenAI | $0.20 | $1.20 | ★★★☆☆ | Budget OpenAI option |
| GPT-5.4 nano | OpenAI | $0.20 | $1.25 | ★★☆☆☆ | Cheapest OpenAI |
| Gemini 3 Flash | $0.50 | $3.00 | ★★★☆☆ | Mid-range, solid | |
| Mistral Large 3 | Mistral | $0.50 | $1.50 | ★★★★☆ | Strong open-weight option |
| GPT-5.4 mini | OpenAI | $0.75 | $4.50 | ★★★☆☆ | OpenAI mid-tier |
| Claude Haiku 4.5 | Anthropic | $1.00 | $5.00 | ★★★★☆ | Fast, capable, quality |
| Gemini 3.1 Pro | $2.00 | $12.00 | ★★★★☆ | Strong reasoning | |
| GPT-5.6 Terra | OpenAI | $2.00 | $12.00 | ★★★★☆ | OpenAI workhorse |
| Claude Opus 5 | Anthropic | $5.00 | $25.00 | ★★★★★ | Premium, best GM-ing |
| GPT-5.6 Sol | OpenAI | $5.00 | $30.00 | ★★★★★ | Premium, best GM-ing |
| Claude Fable 5 | Anthropic | $10.00 | $50.00 | ★★★★★ | Ultimate, top-tier GM |
- Cost: Free if self-hosted, $100/mo for Ollama's cloud tier
- Quality: Depends entirely on the model you run locally (Llama 3.3 70B, Mistral, etc.)
- Caveat: Needs decent hardware; can be slow on CPU-only machines
- Gemini 2.5 Flash-Lite: Best for high-volume, low-stakes narration
- Mistral Small 4: Strong open-source option with good multi-language support
- GPT-5.6 Luna: OpenAI's budget option; cuts costs by 80% over Terra
- Mistral Large 3: Excellent open-weight value; $0.50/$1.50
- Claude Haiku 4.5: Anthropic's budget model, but punches above its weight for narrative
- GPT-5.6 Terra: Good balance of cost and capability
- Claude Opus 5: Anthropic's flagship; excellent at complex narrative reasoning
- GPT-5.6 Sol: OpenAI's frontier model; best for rich, consistent storytelling
- Claude Fable 5: Anthropic's creative tier; designed for narrative tasks
A typical 4-hour TTRPG session with your AI GM Bot might use:
- Input tokens: 50,000–150,000 (system prompt, adventure state, chat history, character sheets)
- Output tokens: 100,000–400,000 (narration, NPC dialogue, scene descriptions)
Here's the estimated cost per session for each tier:
| Tier | Example Model | Cost per 4-hour session |
|---|---|---|
| Free | Local Ollama | $0 |
| Budget | GPT-5.6 Luna, Mistral Small 4 | ~$0.10–$0.50 |
| Mid | Mistral Large 3, Claude Haiku 4.5 | ~$0.50–$2.50 |
| Premium | GPT-5.6 Terra, Gemini Pro | ~$2.50–$6.00 |
| Flagship | GPT-5.6 Sol, Claude Opus 5 | ~$5.00–$12.00 |
| Ultra | Claude Fable 5 | ~$8.00–$20.00 |
Context Caching cuts repeated input tokens by ~50–90%:
- OpenAI: 90% off cached input tokens
- DeepSeek: cache hits as low as $0.15/M tokens
- Groq: 50% off cached input tokens
Batch API offers ~50% off if you can wait 24 hours
Peak pricing matters for DeepSeek: weekday peak hours cost double off-peak
| User Type | Recommendation | Why |
|---|---|---|
| Privacy-first / zero-cost | Ollama local | No data leaves your machine |
| Casual GM | Mistral Small 4 or GPT-5.6 Luna | Great value, good enough |
| Serious GM | Mistral Large 3 or Claude Haiku 4.5 | Strong quality without premium prices |
| Professional / Streamer | GPT-5.6 Sol or Claude Opus 5 | Consistent quality for audience experience |
| Narrative-focused | Claude Fable 5 | Purpose-built for storytelling |
All prices as of August 2026. Check provider pages for updates.