Pricing

Pay-Per-Action

Franklin uses micropayments— you pay per API call, settled instantly in USDC on Base or Solana via the x402 protocol. There are no subscriptions, monthly fees, or rate limits.

Your cost = provider cost + 5%. That's it. Each model call costs fractions of a cent, proportional to the tokens consumed.

Free Tier

Franklin always includes free access to NVIDIA-hosted models with unlimited usage:

• NVIDIA GPT-OSS 120B / 20B
• Llama 3.3 70B
• Qwen 2.5
• DeepSeek R1

Smart Router Cost Optimization

The Smart Router automatically optimizes cost by selecting the cheapest model capable of handling the current task. For simple tasks (file reads, shell commands), it routes to free or low-cost models. For complex reasoning, it routes to premium models.

You can override this by pinning a specific model:

# Use a specific model
franklin -m claude-sonnet-4-20250514

# See all available models and their cost tiers
franklin models

Cost Tracking

Franklin tracks every cent spent. Use these tools to monitor your usage:

In-session

# Inside a Franklin session
/cost

Shows the running cost of the current session, broken down by model.

From the command line

franklin insights

Displays aggregated spending analytics: total cost, cost per model, cost per session, and trends over time.

Web panel

franklin panel

Opens a web dashboard with charts showing spending history, model usage breakdown, and cost projections.

What $1 Gets You

Concrete token budgets for $1 in USDC:

• GPT-4o— ~400K input tokens
• DeepSeek V3— ~7M input tokens
• Gemini Flash— ~13M input tokens
• DALL-E 3— ~20 images
• Exa neural search— ~40 queries
• NVIDIA GPT-OSS— Unlimited (FREE)

Example Session Costs

Approximate costs for common operations (varies by model and context size):

• Simple question— $0.001 – $0.005
• File edit with tool calls— $0.005 – $0.02
• Complex multi-step task— $0.02 – $0.10
• Full coding session (30 min)— $0.10 – $0.50