Documentation
¶
Overview ¶
Package pricingdata is the hand-maintained source of Studio's default pricing table (plan A5): studio/pricing.json is generated from it by studio/internal/pricinggen (go generate ./studio/...), committed, and held equal to JSON's output by studio's TestPricingJSONIsGenerated.
Edit the table here, never the JSON: bump Dated to the day the rates were read, run go generate, commit both.
The rates are the vendors' public per-1M-token list prices in USD for the first-party APIs (not Bedrock, Vertex or Azure resale), read on Dated. Rules the table keeps:
- a model whose price depends on the prompt's length (a long-context tier) is left out: one rate per token kind cannot price it exactly;
- a model whose rate the author was not sure of is left out — an unpriced run reads "unpriced", never a guessed number;
- a rate column the vendor does not publish is omitted, and Studio then prices that token kind at the base rate (input or output);
- cache_write is the 5-minute-TTL write rate (Anthropic: 1.25× the input rate); a 1-hour write bills 2×, which the usage cannot tell apart, so a 1-hour cache write is understated;
- the Gemini rows carry no cached_input (the context-cache rates were not certain): a cached Gemini prompt is priced at the input rate, an overstatement;
- a premium speed tier (Claude Opus 5.5 fast mode, 2× the rates) is not modelled: the usage does not say which speed served a call;
- reasoning tokens bill as output at every vendor listed, so no reasoning column is set;
- globs are path.Match patterns over "provider/model", first match wins, so the most specific entry comes first; exact model ids are preferred over wildcards so a future model never inherits an old price.
Index ¶
Constants ¶
View Source
const Dated = "2026-10-10"
Dated is the day the rates below were read (ISO 8601).
View Source
const Note = "Prices are the vendors' public per-1M-token list rates in USD (first-party APIs), read on the dated day. " +
"model_glob is a path.Match pattern over provider/model; the first match wins. A row with no provider (or provider \"*\") " +
"matches on model_glob alone, any transport (an OpenAI-compatible endpoint behind the openai adapter). " +
"A scripted model no row matches (wefttest/*, weft/runtime/scripted; any other wrapper ships its own row) is an exact $0, badge scripted: a rule in Studio, not a row. A missing cached_input, " +
"cache_write or reasoning rate prices those tokens at the input or output rate. " +
"cache_write is the 5-minute-TTL rate (1.25x input): a 1-hour cache write (2x) is understated. " +
"The Gemini rows have no cached_input: a cached Gemini prompt is priced at the input rate. " +
"Premium speed tiers (Claude Opus 5.5 fast mode) and long-context tiers are not modelled; models with a long-context tier are left out. " +
"This file is also the template for a WEFT_PRICING / --pricing override, which replaces this table whole. " +
"Generated from studio/internal/pricingdata: edit there, then go generate ./studio/..."
Note is the JSON file's own comment (JSON has none of its own).
Variables ¶
View Source
var Models = []Model{
anthropic("claude-fable-5-1", 10, 50, 0.25),
anthropic("claude-mythos-5-1", 10, 50, 0.25),
anthropic("claude-fable-5", 10, 50, 1),
anthropic("claude-mythos-5", 10, 50, 1),
anthropic("claude-opus-5-5", 4, 20, 0.20),
anthropic("claude-opus-5", 5, 25, 0.50),
anthropic("claude-opus-4-8", 5, 25, 0.50),
anthropic("claude-opus-4-7", 5, 25, 0.50),
anthropic("claude-opus-4-6", 5, 25, 0.50),
anthropic("claude-sonnet-5-5", 2, 10, 0.20),
anthropic("claude-sonnet-5", 2, 10, 0.20),
anthropic("claude-sonnet-4-6", 3, 15, 0.30),
anthropic("claude-haiku-4-5", 1, 5, 0.10),
anthropic("claude-haiku-4-5-20*", 1, 5, 0.10),
anthropic("claude-opus-4-5", 5, 25, 0.50),
anthropic("claude-opus-4-5-20*", 5, 25, 0.50),
anthropic("claude-opus-4-1", 15, 75, 1.50),
anthropic("claude-opus-4-1-20*", 15, 75, 1.50),
openai("gpt-5", 1.25, 10, 0.125),
openai("gpt-5-2025-08-07", 1.25, 10, 0.125),
openai("gpt-5-mini", 0.25, 2, 0.025),
openai("gpt-5-mini-2025-08-07", 0.25, 2, 0.025),
openai("gpt-5-nano", 0.05, 0.40, 0.005),
openai("gpt-5-nano-2025-08-07", 0.05, 0.40, 0.005),
openai("gpt-4.1", 2, 8, 0.50),
openai("gpt-4.1-2025-04-14", 2, 8, 0.50),
openai("gpt-4.1-mini", 0.40, 1.60, 0.10),
openai("gpt-4.1-mini-2025-04-14", 0.40, 1.60, 0.10),
openai("gpt-4.1-nano", 0.10, 0.40, 0.025),
openai("gpt-4.1-nano-2025-04-14", 0.10, 0.40, 0.025),
openai("gpt-4o", 2.50, 10, 1.25),
openai("gpt-4o-2024-08-06", 2.50, 10, 1.25),
openai("gpt-4o-2024-11-20", 2.50, 10, 1.25),
openai("gpt-4o-mini", 0.15, 0.60, 0.075),
openai("gpt-4o-mini-2024-07-18", 0.15, 0.60, 0.075),
openai("o3", 2, 8, 0.50),
openai("o3-2025-04-16", 2, 8, 0.50),
openai("o3-mini", 1.10, 4.40, 0.55),
openai("o3-mini-2025-01-31", 1.10, 4.40, 0.55),
openai("o4-mini", 1.10, 4.40, 0.275),
openai("o4-mini-2025-04-16", 1.10, 4.40, 0.275),
openai("o1", 15, 60, 7.50),
openai("o1-2024-12-17", 15, 60, 7.50),
google("gemini-2.5-flash", 0.30, 2.50),
google("gemini-2.5-flash-lite", 0.10, 0.40),
google("gemini-2.0-flash", 0.10, 0.40),
google("gemini-2.0-flash-lite", 0.075, 0.30),
}
Models is the table, most specific first within a family.
Functions ¶
Types ¶
type Model ¶
type Model struct {
Provider string `json:"provider"`
ModelGlob string `json:"model_glob"`
Input float64 `json:"input"`
Output float64 `json:"output"`
CachedInput *float64 `json:"cached_input,omitempty"`
CacheWrite *float64 `json:"cache_write,omitempty"`
Reasoning *float64 `json:"reasoning,omitempty"`
}
Model is one row: a provider, a model glob and its rates per 1M tokens. Nil rates are omitted from the JSON.
Source Files
Click to show internal directories.
Click to hide internal directories.