AI Cost Planner for Games
Plan what AI will cost per month and per player: LLM NPC dialogue, voice, speech recognition, moderation, generated art and 3D, coding agents and localization, with current model prices.
How to use
- Pick a scenario: indie NPC dialogue, voiced mobile companion, voice-driven RPG, studio coding agents, asset pipeline or localization.
- Adjust players, sessions, replies and prompt sizes; choose the main and cheap models and turn caching on or off.
- Add voice, speech recognition and the AI you use to make the game.
- Read the monthly total, the range, cost per DAU, session and reply, and the share of revenue when ARPDAU is set; export Markdown or JSON or copy a share link.
What lowers the bill
Prompt caching of the static prompt, a short history window, a cheap model for simple lines and short replies cut runtime cost the most. For localization and other offline work, the Batch API halves the token price.
The result is a planning estimate: real usage depends on players' behaviour, retries and tokenizer differences. Measure tokens in a playtest and put the real numbers back into the planner.
AI for making the game (per month)
Model prices used (checked 2026-10-13)
| Model | Input | Cached | Output | Source |
|---|---|---|---|---|
| Claude Fable 5.1 | $10 | $0.25 | $50 | Anthropic |
| Claude Opus 5.5 | $4 | $0.2 | $20 | Anthropic |
| Claude Sonnet 5.5 | $2 | $0.1 | $10 | Anthropic |
| Claude Haiku 5.5 (prompts up to 100K) | $0.1 | $0.01 | $0.5 | Anthropic |
| GPT-5.5 | $5 | $0.5 | $30 | OpenAI |
| GPT-5.4 | $2.5 | $0.25 | $15 | OpenAI |
| GPT-5.4 mini | $0.75 | $0.075 | $4.5 | OpenAI |
| GPT-5.4 nano | $0.2 | $0.02 | $1.25 | OpenAI |
| Gemini 3.1 Pro (prompts up to 200K) | $2 | $0.2 | $12 | |
| Gemini 3.8 Flash | $0.75 | $0.075 | $3.75 | |
| Gemini 3.5 Flash-Lite | $0.3 | $0.03 | $2.5 | |
| DeepSeek V4.1 Flash (peak hours) | $0.3 | $0.006 | $1.2 | DeepSeek |
| DeepSeek V4 Pro (peak hours) | $1.32 | $0.044 | $3.96 | DeepSeek |
| Grok 4.3 | $1.25 | $0.2 | $2.5 | xAI |
| Mistral Large | $0.5 | $0.05 | $1.5 | Mistral |
USD per 1M tokens, standard tier. Providers change prices often: check the source before budgeting. Token counts differ between tokenizers, so the same text can cost 10-30% more on some models.
FAQ
- How is the LLM NPC cost calculated?
- Sessions per month = DAU × AI sessions per day × days. Each NPC reply sends the static prompt (system prompt, lore, character), the recent history and the player's line, and returns the reply. With prompt caching the static prompt is written to the cache once per session and read at the cached price on later replies.
- Why does the history setting matter so much?
- Every reply resends the earlier turns. A 20-reply session with 10 turns of history sends far more input tokens than 20 single questions. Summarise old turns or cap the window to keep the cost flat.
- How current are the prices?
- The table shows the date the prices were checked and links to each provider's pricing page. Every price is editable, and a custom model lets you enter your own contract rates.
- What is the cheap model share?
- Many lines (greetings, barks, shop talk) do not need a frontier model. Routing a share of replies to a small model is the biggest lever after caching.
Related tools
-
AI Material Prompt Builder
Compose prompts for game-ready PBR materials and variants.
-
Game Development Prompt Library
120+ prompts for design, code, art, audio, marketing and NPC AI: search, fill, copy.
-
UI Flow + Screen Generator (UI Toolkit / uGUI)
Screen-flow JSON, UXML/USS skeleton, ViewModel stub, localization keys — export bundle.
-
Stable Diffusion Time Estimator
Steps, resolution, batch, it/s guess — wall-clock and throughput.
-
RL Reward Calculator (lite)
Sparse/dense/shaping terms → combined reward trace over toy episode.
-
Diffusion Loss Calculator (lite)
Simple MSE / signal-noise toy loss from numeric inputs.