Usage and pricing
The terminal reports token usage and an estimated API cost after responses.
Use :token for session usage. Account-plan connections display a plan indicator
instead of an estimated API charge; plan limits and credits remain provider-managed.
Read the numbers
| Display | Meaning |
|---|---|
| Context bar | Latest reported request input/output against the built-in context estimate |
| Session tokens | Accumulated usage, including delegated model work |
| API cost | Estimate from the bundled provider/model pricing catalog |
| Account-plan indicator | Account usage link, not an API billing estimate |
Unknown models use a 128,000-token context estimate. The context display is not a live guarantee of the provider’s actual window.
Unknown prices contribute zero to numeric cost totals. Zero means “not priced,” not “free.” Cost budgets cannot account for charges missing from the catalog. Estimates are neither invoices nor guaranteed billing caps.
Snapshot coverage
The embedded USD text-token snapshot was reviewed on October 30, 2026 and covers every bundled default and fallback model choice across ten providers. Discovery can return newer or additional models without bundled prices. Historical entries remain available but are not all newly verified.
Lookup is provider-specific: a direct-provider model does not inherit OpenRouter’s rate, and an unknown Azure deployment does not inherit a public model tariff.
Prompt-length tiers
For supported tiers, the request’s input-token count selects rates for all input and output tokens, not only the tokens above a threshold.
- Grok switches at 200,000 input tokens.
- Direct Gemini Pro switches above 200,000.
- OpenRouter uses its published
min_prompt_tokensboundaries.
Provider-specific assumptions
| Provider | Estimate basis and limits |
|---|---|
| Azure OpenAI | Global Standard reference rates for deployments named like known model IDs; no inference of region, deployment type, or negotiated rates |
| Qwen | International standard list prices, not mainland-China regional or promotional prices; qwen-plus uses the higher thinking-mode output rate |
| DeepSeek | Peak prices, excluding off-peak and holiday discounts, including routed OpenRouter entries |
| Gemini | Paid standard text rates, excluding free tiers |
| Ollama | Bundled local qwen3-coder has no hosted token tariff; hardware, power, hosting, and subscriptions are excluded |
Gemini Argon’s introductory price ends November 30, 2026; Flash 3.8 / 3.6 introductory prices end December 31, 2026. The snapshot does not automatically change on those dates; update it before relying on later estimates.
Other local or cloud IDs do not inherit another provider’s price.
Excluded charges
Usage reports input/output totals without cache-read/write breakdowns. Calculations therefore use uncached input rates; stored cache prices are metadata. Tool fees, tax, residency, batch discounts, premium-speed modes, and routing adjustments are not included.
For billing decisions, compare Codai’s estimates with your provider dashboard.
For SDK hosts
EmbeddedPricingCatalog.Find(provider, model) returns no entry for unknown prices.
ModelPricing.InputTokenTiers exposes ModelPricingTier entries with inclusive
MinInputTokens thresholds and per-token USD input/output rates.
Calculate selects the highest applicable threshold.
PricingDocument.CurrentModels exposes the reviewed provider-qualified overrides.
Loading overrides preserves historical context and capability metadata when not replaced.
Pricing sources
OpenAI · Azure · Anthropic · Gemini · DeepSeek · Qwen · Mistral · xAI · OpenRouter · Ollama .