Compute Credits: Pay Only For What You Use
April 4, 2026 · By Ashim Nandi
ATOM uses usage-based compute credits instead of flat subscriptions. Every AI operation costs credits proportional to the compute it consumes. Light months cost less. Research-heavy months cost more. You never pay for access you do not use. This model aligns cost with value and eliminates the subscription penalty for traders who step back from the market.
Why Usage-Based
The subscription model has a structural problem. It charges the same whether you trade every day or take a month off.
A trader who runs deep research in January, executes heavily in February, and steps away in March to reset pays the same flat fee all three months. Two of those months represent a fair exchange. One does not.
Usage-based pricing fixes this.
| Month | Activity | Flat Subscription | Usage-Based |
|---|---|---|---|
| January | Heavy research, 200 tool calls | $69 | ~$12 |
| February | Active trading, 800 tool calls | $69 | ~$38 |
| March | Away from market, 10 tool calls | $69 | ~$1 |
| Total | $207 | ~$51 |
The numbers above are illustrative, but the principle holds. Usage-based pricing rewards disciplined, selective trading. It does not penalize you for waiting. In systematic trading, the ability to do nothing and pay nothing for that discipline is itself valuable.
How Credits Work
Every AI operation in ATOM consumes compute credits. The cost varies by two factors: the model used and the operation type.
Model cost. Larger, more capable models cost more per token. Claude Haiku processing a simple market data query costs a fraction of what Claude Opus costs for a deep strategy analysis. You control this by selecting your model.
Operation complexity. A quick price lookup costs less than a full Monte Carlo simulation across 10,000 equity curve permutations. The compute required determines the credit cost.
Approximate Costs by Operation Type
| Operation | Typical Credit Cost | Notes |
|---|---|---|
| Market data query | Low | Price, volume, basic fundamentals |
| Portfolio summary | Low | Current holdings, P&L |
| Chat message (fast model) | Low | Haiku, Llama, simple queries |
| Chat message (balanced model) | Medium | Sonnet, GPT-4o-mini |
| Risk analysis | Medium | Portfolio heat, VaR, drawdown |
| G-Score calculation | Medium | Strategy quality assessment |
| Backtest (single run) | Medium-High | Walk-forward, overfitting checks |
| Chat message (reasoning model) | High | Opus, deep analysis |
| Monte Carlo simulation | High | Thousands of equity curve simulations |
| Full strategy evaluation | High | Multi-factor, regime-segmented analysis |
Credits are deducted in real time. Your balance updates after each operation. Low-balance warnings trigger at $1 remaining and again at $0.10. You never hit a surprise zero.
Model Pricing
ATOM offers nine LLM models. Each sits at a different point on the cost-capability spectrum. You choose the right model for the task.
| Model | Cost Tier | Best For |
|---|---|---|
| Claude Haiku | Low | Quick lookups, simple questions, data retrieval |
| Llama models | Low | Fast iteration, bulk queries |
| GPT-4o-mini | Low-Medium | Balanced everyday tasks |
| Claude Sonnet | Medium | Analysis, strategy discussion, risk review |
| GPT-4o | Medium-High | Complex reasoning, multi-step analysis |
| Claude Opus | High | Deep research, novel strategy development, nuanced judgment |
The model selector is available in both the web interface and the CLI. Switch models mid-conversation based on what the task demands. Use Haiku to scan 50 instruments, then switch to Opus to deeply analyze the top three candidates. This is how you control costs without sacrificing capability.
The model is the interface. The domain layer is the product. Switching from Opus to Haiku changes how you interact with ATOM's 187 services. It does not change the services themselves. G-Score, risk calculations, backtests, and broker execution are computed by the domain layer regardless of which model you chose.
BYOK: Bring Your Own Key
Power users who already have API keys from Anthropic, OpenAI, or other providers can plug them directly into ATOM. This is BYOK (Bring Your Own Key).
With BYOK:
- Zero markup on LLM inference. You pay your provider's rates directly. ATOM adds nothing.
- Credits only consumed for domain services. Risk checks, market data, broker execution, backtests, and other domain operations still cost credits. The LLM inference does not.
- Full model access. Use any model your API key supports, including models not in ATOM's default selector.
BYOK is designed for high-volume users who want maximum control over their LLM costs. It separates the two cost components cleanly: LLM inference (your key, your cost) and domain services (ATOM credits).
Accepted Currencies
Credits can be purchased with five currencies. All on-chain payments are verified before credits are issued.
| Currency | Type | Verification |
|---|---|---|
| USDC | Stablecoin | On-chain transaction verification |
| USDT | Stablecoin | On-chain transaction verification |
| SOL | Solana native | On-chain transaction verification |
| PYUSD | PayPal stablecoin | On-chain transaction verification |
| OSR | System R governance token | On-chain verification + 50% bonus applied |
Fiat payments via Stripe are also supported for users who prefer credit card or bank transfer.
OSR Token Bonus
OSR is System R's governance token. When used to purchase compute credits, OSR holders receive a 50% bonus.
Buy $10 of credits with OSR. Get $15 worth.
This is the primary utility of the OSR token beyond governance. It is not a speculative mechanic. It is a straightforward discount for ecosystem participants.
| Purchase Amount (OSR) | Credit Value | Bonus |
|---|---|---|
| $10 | $15 | +$5 (50%) |
| $50 | $75 | +$25 (50%) |
| $100 | $150 | +$50 (50%) |
The bonus applies at purchase time. Once credits are in your account, they spend at the same rate regardless of how they were purchased. Presale participants receive an additional 20% discount on OSR token price, which compounds with the credit bonus.
Free Tier
Every new account receives $5 in free compute credits. No credit card required.
$5 is enough to:
- Run 10-20 chat conversations with a balanced model
- Execute several G-Score calculations
- Scan the market across multiple asset classes
- Run a basic backtest
- Explore the full tool catalog
The free tier exists so you can evaluate ATOM's domain layer with real data before committing money. There is no time limit on the free credits. If you sign up, run a few analyses, and come back three months later, the remaining credits are still there.
For Agents: Dual-Species Pricing
ATOM is a dual-species platform. Human traders and AI agents are both first-class users with separate pricing models.
| User Type | Access | Billing |
|---|---|---|
| Human traders | Chat interface, CLI | Compute credits (any accepted currency) |
| AI agents | SDK, MCP tools | Per-call USDC (programmatic) |
AI agents authenticate with API keys and are billed per MCP tool invocation. Pricing varies by tool category. Market data calls cost less than execution calls. Risk analysis sits in between.
Agent billing is designed for programmatic access at scale. A research agent making hundreds of tool calls per day gets granular cost tracking per call. Usage reports break down costs by tool category, timestamp, and agent ID.
For developers building agents on ATOM, the economics favor agents that are selective and intelligent about tool usage. An agent that calls market_data on 500 instruments before narrowing to 5 candidates costs more than one that uses smart filtering first. Good agent design and good economics align.
FAQ
How do compute credits compare to a flat $69/month subscription?
It depends on your usage pattern. A trader who uses ATOM daily for research, analysis, and execution might spend $30-60 per month in credits. A trader who uses it a few times per week might spend $10-20. A trader who takes a month off spends near zero. The break-even point versus a flat subscription depends on how heavily and consistently you use the platform. For most individual traders, usage-based pricing costs less.
Can credits expire?
No. Credits do not expire. Purchase them when convenient, use them when needed. This is especially important for traders who follow seasonal strategies or take extended breaks from the market.
What happens when my credits run out?
ATOM warns you at $1 remaining and again at $0.10. When credits hit zero, AI operations pause. You can still log in, view your portfolio, and access your workspace. You just cannot make new AI-powered requests until you add credits. No overdraft, no surprise charges.
Is BYOK available on the free tier?
Yes. You can plug in your own API keys from day one. The $5 free credits cover domain service costs (risk checks, market data, backtests). LLM inference costs go directly to your API provider. This is an efficient way to explore ATOM if you already have API keys from Anthropic or OpenAI.