Pricing and Usage
How AVCodex usage-based pricing works, and how to keep AI costs in check on integrator-scale traffic.
Understanding how AI pricing works helps you build better agents and control costs. This guide explains the pricing model and gives you tools to estimate and optimize your spend.
We get asked about pricing a lot, so we will be straight about it.
We want every AV team to have access to the best AI models available. Not watered-down versions. Not artificially capped options. The actual best models from OpenAI, Anthropic, and Google.
These models have real costs. Running GPT-5 costs more than running GPT-5 Nano. Claude Opus 4 costs more than Claude 3.5 Haiku. If we charged a flat rate, we would have to do one of three things:
- Restrict access to premium models, forcing everyone onto cheaper ones.
- Charge everyone the premium price, putting AVCodex out of reach for many integrators.
- Lose money, which means we cannot keep building tools for the AV industry.
Usage-based pricing solves it. It means:
- Cost-conscious teams can stand up high-volume room support agents on fast, cheap models like Gemini 2.5 Flash or GPT-5 Nano and keep monthly costs minimal.
- Quality-focused teams can pull in the heaviest reasoning models (o3, Claude Opus 4) for RFP responses, design engineering, and deep troubleshooting when the work is worth it.
- Smart teams mix and match. Cheap models for triage. Premium models for the hard problems.
This is not about being greedy. It is about running a business that can keep delivering best-in-class AI to every type of AV team, at every budget level.
Your plan includes usage#
Every AVCodex plan includes a monthly usage allowance:
| Plan | Monthly price | Included usage |
|---|---|---|
| Builder | $29/mo | $10 |
| Studio | $99/mo | $30 |
| Studio Pro | $299/mo | $100 |
Stay within your allowance and you pay nothing extra. Need more? Buy additional credits in advance.
What counts as usage#
Usage is measured in tokens, the units AI models use to process text. Roughly:
- 1 token is about 4 characters or three-quarters of a word.
- A typical message uses 500 to 1,000 tokens total (input plus output).
Different models have different per-token costs. Premium models cost more, efficient models cost less.
Want to dig in? See the Understanding Tokens guide with an interactive tokenizer.
Use the calculator in the dashboard to estimate your monthly AI costs based on expected traffic.
How different models stack up on price. Pick based on the work the agent is doing.
Most cost-effective models#
These models give strong quality at the lowest prices.
Premium models#
When you need the best reasoning quality (think a senior design engineer agent reviewing a multi-room AV-over-IP build), these deliver.
1. Match the model to the task#
Do not run a senior reasoning model for FAQ traffic.
| Task type | Recommended approach |
|---|---|
| Quick Q&A, room support | Gemini 2.5 Flash or GPT-5 Nano |
| General conversation, internal SOP help | GPT-5 Mini or Claude 3.5 Haiku |
| Content creation (project narratives, SOWs, drafts) | GPT-5 or Claude Sonnet 4 |
| Complex analysis (design engineering, RFP scoring) | o3 or Claude Opus 4 |
| Long documents (multi-room programming guides, AS-builts) | Gemini 2.5 Pro (best value for 1M context) |
2. Optimize your prompts#
Shorter, clearer prompts use fewer tokens.
- Be specific about what you want.
- Cut unneeded context from system prompts.
- Use tight instructions instead of verbose ones.
3. Watch response length#
If your agent does not need long answers (a tech in a closet wants the fix in two sentences), tell the model to be concise. Output tokens cost more than input tokens.
4. Use the right context window#
Models with large context windows (Gemini 2.5 Pro at 1M tokens) shine on long documents. If your agent only handles short Q&A on touch panel issues, you do not need to pay for that capacity.
Example 1: Room support agent#
Use case: Answering touch panel and routine room questions through a QR code in every conference room.
Recommended: Gemini 2.5 Flash ($0.25/M avg).
Monthly cost for 10,000 messages: about $2.50.
Example 2: Programmer assistant#
Use case: Helping Crestron and Q-SYS programmers draft modules, comments, and code reviews.
Recommended: GPT-5 ($16/M avg).
Monthly cost for 5,000 messages: about $40.
Example 3: RFP and spec analyzer#
Use case: Reviewing 60-page RFPs, extracting requirements, summarizing scope.
Recommended: Gemini 2.5 Pro ($4/M avg, 1M context).
Monthly cost for 1,000 documents: about $20.
Example 4: Design engineering assistant#
Use case: Deep multi-step reasoning on complex AV-over-IP designs, signal flow validation, codec calculations.
Recommended: o3 ($33/M avg).
Monthly cost for 2,000 queries: about $33.
You can track usage in real time:
- Open Settings > Billing in your AVCodex dashboard.
- View your current credit balance and usage history.
- Buy additional credits when you need them.
We notify you when credits are running low, so you have time to top up before your agent pauses.
What happens if I run out of credits?#
Your agent's AI functionality pauses until you purchase more credits. We send notifications as your balance drops, so you are not caught off guard. Top up anytime from billing settings.
How do I add more credits?#
Go to Settings > Billing and buy credits in advance. Credits are prepaid. No surprise bills, no overages. You control exactly what you spend.
Do unused credits roll over?#
No, included usage does not roll over month to month. This keeps pricing predictable.
Why do output tokens cost more than input tokens?#
That reflects how the underlying AI providers price their APIs. Generating new text (output) takes more compute than processing existing text (input).
Can I change models mid-conversation?#
An agent uses one model at a time, but you can change it anytime in Build settings. The change applies to new conversations.
Today, usage billing covers chat interactions and tool calls. We are expanding it to unlock more capability for every team:
Voice interactions#
Real-time voice conversations are expensive. Adding voice to usage billing lets us open the feature to everyone (not only specific tiers). You pay for what you use.
Knowledge source embeddings#
Bulk uploads of knowledge sources are limited today. Adding embedding costs to usage billing lets us optimize for large-scale uploads. You will be able to load hundreds or thousands of documents (every Biamp datasheet, every Crestron module, every commissioning guide) into your agent's knowledge base.
Image and video generation#
AI-generated images and video are resource-intensive (video generation runs $3 to $6 per video). Usage billing lets us offer these tools to every plan. Generate marketing visuals, room walkthroughs, product demos. You only pay for what you create.
Billing and invoices#
The Settings > Billing page will show detailed usage invoices, so you can see exactly where the spend goes and track it over time.
We are building toward every AI capability being available to every team, with transparent, pay-for-what-you-use pricing.
If you are not sure which model fits, see Choosing a Model for recommendations by scenario.
You can read billing and usage information through code using the AVCodex MCP Server.
Available tools#
| Tool | Description |
|---|---|
get_credits |
View current credit balance. |
get_subscription |
Get subscription tier and details. |
get_usage |
Get a detailed usage breakdown. |
Example prompts#
"How many credits do I have left?"
"What is my current subscription plan?"
"Show me my usage breakdown for this month"Workflow: usage monitoring#
You: Show me my current billing status
AI: [Calls get_credits, get_subscription]
Subscription: Studio ($99/mo)
Credit Balance: $18.50 remaining
Monthly Allowance: $30 included
Usage this month:
- GPT-5: $8.20 (410K tokens)
- Claude Sonnet 4: $2.30 (115K tokens)
- Gemini Flash: $1.00 (4M tokens)Note: MCP server access requires a Builder plan or higher. See the MCP Setup Guide to get started.
*AVCodex · Your AV expertise. Amplified by AI.*