diff --git a/README.md b/README.md
index 34bfb83..500f14f 100644
--- a/README.md
+++ b/README.md
@@ -132,7 +132,74 @@ To reset the key in your browser: open DevTools (F12) > Application > Local Stor
 
 ## Cost Calculation
 
-Costs come directly from LibreChat's `transactions` collection. LibreChat stores `tokenValue = rawAmount x rate` where rate is USD per 1M tokens. The dashboard reads these pre-calculated values — it does not independently calculate pricing.
+### How it works
+
+The dashboard does **not** calculate costs itself. Costs come directly from LibreChat's MongoDB `transactions` collection, which records every API call.
+
+When a user sends a message, LibreChat:
+1. Records the token usage in the `transactions` collection with fields like `rawAmount` (token count, stored as negative), `tokenType` (prompt or completion), and `model`
+2. Looks up the model's pricing from its internal pricing table (`api/models/tx.js`)
+3. Calculates `tokenValue = rawAmount × rate` where rate is USD per 1M tokens
+4. Stores `tokenValue` alongside the transaction
+
+The dashboard reads `tokenValue` and converts to USD: **`cost = Math.abs(tokenValue) / 1,000,000`**
+
+This means costs reflect the rate that was active *at the time of the API call*. If pricing changes, old transactions keep their original cost (which is correct — they were billed at the old rate).
+
+### Where pricing lives
+
+The pricing table is in LibreChat's codebase at `api/models/tx.js`. It uses a longest-match pattern strategy — e.g. `claude-sonnet-4-6` matches before the generic `claude-` fallback.
+
+### Agent transactions
+
+When a user interacts with an Agent, LibreChat records `model` as `agent_XYZ` (the agent's ID). The dashboard resolves these to the underlying LLM model via the `agents` collection (cached, refreshed every 5 minutes).
+
+### Verified pricing (March 2026)
+
+The following rates from LibreChat's `tx.js` have been validated against official provider pricing pages:
+
+**Anthropic Claude** (source: platform.claude.com/docs/en/about-claude/pricing)
+
+| Model | Input ($/1M tokens) | Output ($/1M tokens) | Verified |
+|-------|--------------------:|---------------------:|----------|
+| claude-opus-4-6 | 5.00 | 25.00 | Yes |
+| claude-sonnet-4-6 | 3.00 | 15.00 | Yes |
+| claude-sonnet-4 | 3.00 | 15.00 | Yes |
+| claude-opus-4-5 | 5.00 | 25.00 | Yes |
+| claude-opus-4 | 15.00 | 75.00 | Yes |
+| claude-haiku-4-5 | 1.00 | 5.00 | Yes |
+| claude-3-5-haiku | 0.80 | 4.00 | Yes |
+| claude-3-opus | 15.00 | 75.00 | Yes |
+| claude-3-haiku | 0.25 | 1.25 | Yes |
+
+**OpenAI** (source: openai.com/api/pricing)
+
+| Model | Input ($/1M tokens) | Output ($/1M tokens) | Verified |
+|-------|--------------------:|---------------------:|----------|
+| gpt-4o | 2.50 | 10.00 | Yes |
+| gpt-4o-mini | 0.15 | 0.60 | Yes |
+| gpt-4.1 | 2.00 | 8.00 | Yes |
+| gpt-4.1-mini | 0.40 | 1.60 | Yes |
+| gpt-4.1-nano | 0.10 | 0.40 | Yes |
+| o1 | 15.00 | 60.00 | Yes |
+| o1-mini | 1.10 | 4.40 | Mismatch* |
+| o3 | 2.00 | 8.00 | Yes |
+| o3-mini | 1.10 | 4.40 | Yes |
+| o4-mini | 1.10 | 4.40 | Yes |
+
+*o1-mini: tx.js has $1.10/$4.40, current OpenAI pricing shows $3/$12. May reflect an older rate. Only affects you if using o1-mini.
+
+**Google Gemini** (source: ai.google.dev/gemini-api/docs/pricing)
+
+| Model | Input ($/1M tokens) | Output ($/1M tokens) | Verified |
+|-------|--------------------:|---------------------:|----------|
+| gemini-2.5-pro | 1.25 | 10.00 | Yes |
+| gemini-2.5-flash | 0.30 | 2.50 | Yes |
+| gemini-2.0-flash | 0.10 | 0.40 | Yes |
+| gemini-2.0-flash-lite | 0.075 | 0.30 | Yes |
+| gemini-1.5-flash | 0.15 | 0.60 | Deprecated |
+
+Note: Gemini 2.5 Pro has tiered pricing — prompts over 200K tokens cost $2.50/$15.00. tx.js uses the base tier which applies to most usage.
 
 ## API Endpoints