[awf] Calculate and expose AI Credits alongside effective tokens in the API proxy

[lpcox]

Summary

The API proxy currently tracks effective tokens using configurable per-model multipliers (unitless weights). This proposal adds a parallel AI Credits calculation using GitHub's published per-token pricing, giving workflows and operators a direct dollar-cost metric (1 AI credit = $0.01 USD).

Background

GitHub Copilot plans include AI Credits allowances. When usage exceeds allowances, billing is based on per-token rates published at:

• https://docs.github.com/en/copilot/reference/copilot-billing/models-and-pricing
• https://models.dev (cross-provider reference)

The API proxy already intercepts every request/response and extracts token counts from usage metadata. Adding credits calculation is a natural extension.

Proposed Pricing Table (per 1M tokens)

Model	Input	Cached Input	Cache Write	Output
OpenAI
GPT-5 mini	$0.25	$0.025	—	$2.00
GPT-5.2 / GPT-5.2-Codex / GPT-5.3-Codex	$1.75	$0.175	—	$14.00
GPT-5.4	$2.50	$0.25	—	$15.00
GPT-5.4 mini	$0.75	$0.075	—	$4.50
GPT-5.4 nano	$0.20	$0.02	—	$1.25
GPT-5.5	$5.00	$0.50	—	$30.00
Anthropic
Claude Haiku 4.5	$1.00	$0.10	$1.25	$5.00
Claude Sonnet 4/4.5/4.6	$3.00	$0.30	$3.75	$15.00
Claude Opus 4.5/4.6/4.7/4.8	$5.00	$0.50	$6.25	$25.00
Google
Gemini 2.5 Pro	$1.25	$0.125	—	$10.00
Gemini 3 Flash	$0.50	$0.05	—	$3.00
Gemini 3.1 Pro	$2.00	$0.20	—	$12.00
Gemini 3.5 Flash	$1.50	$0.15	—	$9.00
Other
MAI-Code-1-Flash	$0.75	$0.075	—	$4.50
Raptor mini	$0.25	$0.025	—	$2.00

Proposed Implementation

1. Pricing data file: containers/api-proxy/ai-credits-pricing.js

// Per-model pricing in dollars per 1M tokens
// Source: https://docs.github.com/en/copilot/reference/copilot-billing/models-and-pricing
module.exports = {
  'gpt-5-mini':        { input: 0.25,  cachedInput: 0.025,  cacheWrite: null, output: 2.00 },
  'gpt-5.2':           { input: 1.75,  cachedInput: 0.175,  cacheWrite: null, output: 14.00 },
  'gpt-5.2-codex':     { input: 1.75,  cachedInput: 0.175,  cacheWrite: null, output: 14.00 },
  'gpt-5.3-codex':     { input: 1.75,  cachedInput: 0.175,  cacheWrite: null, output: 14.00 },
  'gpt-5.4':           { input: 2.50,  cachedInput: 0.25,   cacheWrite: null, output: 15.00 },
  'gpt-5.4-mini':      { input: 0.75,  cachedInput: 0.075,  cacheWrite: null, output: 4.50 },
  'gpt-5.4-nano':      { input: 0.20,  cachedInput: 0.02,   cacheWrite: null, output: 1.25 },
  'gpt-5.5':           { input: 5.00,  cachedInput: 0.50,   cacheWrite: null, output: 30.00 },
  'claude-haiku-4-5':  { input: 1.00,  cachedInput: 0.10,   cacheWrite: 1.25, output: 5.00 },
  'claude-sonnet-4':   { input: 3.00,  cachedInput: 0.30,   cacheWrite: 3.75, output: 15.00 },
  'claude-sonnet-4-5': { input: 3.00,  cachedInput: 0.30,   cacheWrite: 3.75, output: 15.00 },
  'claude-sonnet-4-6': { input: 3.00,  cachedInput: 0.30,   cacheWrite: 3.75, output: 15.00 },
  'claude-opus-4-5':   { input: 5.00,  cachedInput: 0.50,   cacheWrite: 6.25, output: 25.00 },
  'claude-opus-4-6':   { input: 5.00,  cachedInput: 0.50,   cacheWrite: 6.25, output: 25.00 },
  'claude-opus-4-7':   { input: 5.00,  cachedInput: 0.50,   cacheWrite: 6.25, output: 25.00 },
  'claude-opus-4-8':   { input: 5.00,  cachedInput: 0.50,   cacheWrite: 6.25, output: 25.00 },
  'gemini-2.5-pro':    { input: 1.25,  cachedInput: 0.125,  cacheWrite: null, output: 10.00 },
  'gemini-3-flash':    { input: 0.50,  cachedInput: 0.05,   cacheWrite: null, output: 3.00 },
  'gemini-3.1-pro':    { input: 2.00,  cachedInput: 0.20,   cacheWrite: null, output: 12.00 },
  'gemini-3.5-flash':  { input: 1.50,  cachedInput: 0.15,   cacheWrite: null, output: 9.00 },
  'mai-code-1-flash':  { input: 0.75,  cachedInput: 0.075,  cacheWrite: null, output: 4.50 },
  'raptor-mini':       { input: 0.25,  cachedInput: 0.025,  cacheWrite: null, output: 2.00 },
};

2. Credits calculation module: containers/api-proxy/ai-credits-calculator.js

Parallels the effective-token-guard pattern:

• On each response, look up the model's pricing
• Calculate: credits = (inputTokens × inputRate + cachedTokens × cachedRate + cacheWriteTokens × writeRate + outputTokens × outputRate) / 1_000_000 / 0.01
• The division by $0.01 converts dollars to AI credits (1 credit = $0.01)
• Accumulate a running total per session

For unknown models, log a warning and skip credits calculation (don't block the request).

3. Expose via management endpoint

Add ai_credits to the /management response alongside effective_tokens:

{
  "effective_tokens": { ... },
  "ai_credits": {
    "total": 4.23,
    "by_model": {
      "claude-sonnet-4-6": { "input_credits": 1.5, "output_credits": 2.73, "total": 4.23 }
    }
  }
}

4. Expose via environment/logs

• Log ai_credits_this_response alongside existing effectiveTokensThisResponse in the per-request log
• Optionally expose AWF_AI_CREDITS_USED env var (or file) for the agent to read

5. Optional: credits-based budget guard

Similar to --max-effective-tokens, add --max-ai-credits <N> to halt a run when it exceeds N credits.

Design Considerations

1. Pricing updates: The pricing table is static at build time. When GitHub publishes new rates, bump the file. Consider a future --ai-credits-pricing-file override for custom/BYOK pricing.

2. Model name resolution: The proxy already resolves model aliases (e.g. COPILOT_MODEL). Credits lookup should use the resolved model name.

3. Cache write tokens: Anthropic reports cache_creation_input_tokens separately from input_tokens. The proxy already extracts this for token tracking — reuse it for credits.

4. No blocking by default: Credits calculation is purely observational unless --max-ai-credits is set.

5. Relationship to effective tokens: These are complementary, not replacements. Effective tokens are unitless weights for relative cost control; AI credits are absolute dollar-denominated costs.

References

• Effective token guard: containers/api-proxy/guards/effective-token-guard.js
• Token usage extraction: containers/api-proxy/proxy-request.js (lines ~500-520)
• Management endpoint: containers/api-proxy/management.js
• GitHub pricing docs: https://docs.github.com/en/copilot/reference/copilot-billing/models-and-pricing