diff --git a/rust/crates/api/src/types.rs b/rust/crates/api/src/types.rs index a2d01d82a8..1587ffc4a2 100644 --- a/rust/crates/api/src/types.rs +++ b/rust/crates/api/src/types.rs @@ -1,6 +1,6 @@ use std::collections::BTreeMap; -use runtime::{pricing_for_model, TokenUsage, UsageCostEstimate}; +use runtime::{pricing_for_request, TokenUsage, UsageCostEstimate}; use serde::{Deserialize, Serialize}; use serde_json::Value; @@ -213,7 +213,9 @@ impl Usage { #[must_use] pub fn estimated_cost_usd(&self, model: &str) -> UsageCostEstimate { let usage = self.token_usage(); - pricing_for_model(model).map_or_else( + let prompt_tokens = + self.input_tokens + self.cache_creation_input_tokens + self.cache_read_input_tokens; + pricing_for_request(model, prompt_tokens).map_or_else( || usage.estimate_cost_usd(), |pricing| usage.estimate_cost_usd_with_pricing(pricing), ) diff --git a/rust/crates/runtime/src/lib.rs b/rust/crates/runtime/src/lib.rs index 5af860ce0f..b138ec8406 100644 --- a/rust/crates/runtime/src/lib.rs +++ b/rust/crates/runtime/src/lib.rs @@ -200,7 +200,8 @@ pub use task_registry::{LaneBoard, LaneBoardEntry, LaneFreshness, LaneHeartbeat} #[cfg(test)] pub use trust_resolver::{TrustConfig, TrustDecision, TrustEvent, TrustPolicy, TrustResolver}; pub use usage::{ - format_usd, pricing_for_model, ModelPricing, TokenUsage, UsageCostEstimate, UsageTracker, + format_usd, pricing_for_model, pricing_for_request, ModelPricing, TokenUsage, + UsageCostEstimate, UsageTracker, }; pub use worker_boot::{ Worker, WorkerEvent, WorkerEventKind, WorkerEventPayload, WorkerFailure, WorkerFailureKind, diff --git a/rust/crates/runtime/src/usage.rs b/rust/crates/runtime/src/usage.rs index 19b684e03d..cf5ddab69f 100644 --- a/rust/crates/runtime/src/usage.rs +++ b/rust/crates/runtime/src/usage.rs @@ -25,6 +25,10 @@ impl ModelPricing { ) } + const fn from_rates([input, output, cache_creation, cache_read]: [f64; 4]) -> Self { + Self::per_million(input, output, cache_creation, cache_read) + } + const fn per_million(input: f64, output: f64, cache_creation: f64, cache_read: f64) -> Self { Self { input_cost_per_million: input, @@ -83,8 +87,8 @@ pub fn pricing_for_model(model: &str) -> Option { return Some(ModelPricing::per_million(0.8, 4.0, 1.0, 0.08)); } return Some(match family_version(&normalized, "haiku") { - // Haiku 5.5 bills prompts over 100K tokens at 5x; usage totals are not - // tracked per request, so estimates use the base tier. + // Haiku 5.5 bills prompts over 100K tokens at 5x; aggregated usage uses the + // base tier, and `pricing_for_request` applies the higher one per request. Some((5, 5)) => ModelPricing::per_million(0.1, 0.5, 0.125, 0.01), _ => ModelPricing::per_million(1.0, 5.0, 1.25, 0.1), }); @@ -134,9 +138,9 @@ const GROK_BUILD: [f64; 4] = [1.0, 2.0, 1.0, 0.2]; /// OpenAI-compatible providers, keyed by model id. /// /// An id also matches its dated snapshots and `-latest` alias (e.g. `gpt-4.1-2025-04-14`), -/// so a specific snapshot must precede its base id. Long-context and higher input tiers -/// are not modeled because usage is aggregated across requests. Cache writes without a -/// separate charge, and models without a cached-input discount, use the input rate. +/// so a specific snapshot must precede its base id. Long-context tiers live in +/// [`LONG_CONTEXT_PRICING`]. Cache writes without a separate charge, and models without a +/// cached-input discount, use the input rate. const OPENAI_COMPATIBLE_PRICING: &[(&str, [f64; 4])] = &[ // OpenAI standard tier (). ("gpt-6-astra", [10.0, 50.0, 12.5, 1.0]), @@ -147,14 +151,19 @@ const OPENAI_COMPATIBLE_PRICING: &[(&str, [f64; 4])] = &[ ("gpt-5.6-terra", [2.0, 12.0, 2.5, 0.2]), ("gpt-5.6-luna", [0.2, 1.2, 0.25, 0.02]), ("gpt-5.5", [5.0, 30.0, 5.0, 0.5]), + ("gpt-5.5-pro", [30.0, 180.0, 30.0, 30.0]), ("gpt-5.4", [2.5, 15.0, 2.5, 0.25]), + ("gpt-5.4-pro", [30.0, 180.0, 30.0, 30.0]), ("gpt-5.4-mini", [0.75, 4.5, 0.75, 0.075]), ("gpt-5.4-nano", [0.2, 1.25, 0.2, 0.02]), + ("gpt-5.3-codex", [1.75, 14.0, 1.75, 0.175]), ("gpt-5.2", [1.75, 14.0, 1.75, 0.175]), + ("gpt-5.2-pro", [21.0, 168.0, 21.0, 21.0]), ("gpt-5.1", [1.25, 10.0, 1.25, 0.125]), ("gpt-5", [1.25, 10.0, 1.25, 0.125]), ("gpt-5-mini", [0.25, 2.0, 0.25, 0.025]), ("gpt-5-nano", [0.05, 0.4, 0.05, 0.005]), + ("gpt-5-pro", [15.0, 120.0, 15.0, 15.0]), ("gpt-4.1", [2.0, 8.0, 2.0, 0.5]), ("gpt-4.1-mini", [0.4, 1.6, 0.4, 0.1]), ("gpt-4.1-nano", [0.1, 0.4, 0.1, 0.025]), @@ -162,9 +171,24 @@ const OPENAI_COMPATIBLE_PRICING: &[(&str, [f64; 4])] = &[ ("gpt-4o", [2.5, 10.0, 2.5, 1.25]), ("gpt-4o-mini", [0.15, 0.6, 0.15, 0.075]), ("o1", [15.0, 60.0, 15.0, 7.5]), + ("o1-pro", [150.0, 600.0, 150.0, 150.0]), ("o3", [2.0, 8.0, 2.0, 0.5]), + ("o3-pro", [20.0, 80.0, 20.0, 20.0]), ("o3-mini", [1.1, 4.4, 1.1, 0.55]), ("o4-mini", [1.1, 4.4, 1.1, 0.275]), + // Google Gemini API paid tier (), + // reachable through its OpenAI-compatible endpoint and gateways. Context caching + // bills storage per hour rather than a write premium, so writes use the input rate. + ("gemini-4-argon", [2.0, 10.0, 2.0, 0.1]), + ("gemini-3.8-flash", [0.75, 3.75, 0.75, 0.075]), + ("gemini-3.6-flash", [0.75, 3.75, 0.75, 0.075]), + ("gemini-3.5-flash-lite", [0.3, 2.5, 0.3, 0.03]), + ("gemini-3.1-flash-lite", [0.25, 1.5, 0.25, 0.025]), + ("gemini-3.1-pro-preview", [2.0, 12.0, 2.0, 0.2]), + ("gemini-3-flash-preview", [0.5, 3.0, 0.5, 0.05]), + ("gemini-2.5-pro", [1.25, 10.0, 1.25, 0.125]), + ("gemini-2.5-flash", [0.3, 2.5, 0.3, 0.03]), + ("gemini-2.5-flash-lite", [0.1, 0.4, 0.1, 0.01]), // xAI (). ("grok-4.7", [2.0, 6.0, 2.0, 0.5]), ("grok-4.6", [2.0, 6.0, 2.0, 0.5]), @@ -189,9 +213,52 @@ const OPENAI_COMPATIBLE_PRICING: &[(&str, [f64; 4])] = &[ ("kimi-k2.5", [0.574, 3.011, 0.574, 0.1148]), ]; +/// Rates that replace the base tier once a single request's prompt (input plus cache +/// reads and writes) exceeds the threshold, keyed like [`OPENAI_COMPATIBLE_PRICING`]. +const LONG_CONTEXT_PRICING: &[(&str, (u32, [f64; 4]))] = &[ + ("gpt-6-astra", (272_000, [20.0, 75.0, 25.0, 2.0])), + ("gpt-6.1-sol", (272_000, [4.0, 15.0, 5.0, 0.2])), + ("gpt-6-sol", (272_000, [4.0, 15.0, 5.0, 0.4])), + ("gpt-6-luna", (272_000, [0.2, 0.75, 0.25, 0.02])), + ("gpt-5.6-sol", (272_000, [8.0, 30.0, 10.0, 0.8])), + ("gpt-5.6-terra", (272_000, [4.0, 18.0, 5.0, 0.4])), + ("gpt-5.6-luna", (272_000, [0.4, 1.8, 0.5, 0.04])), + ("gpt-5.5", (272_000, [10.0, 45.0, 10.0, 1.0])), + ("gpt-5.5-pro", (272_000, [60.0, 270.0, 60.0, 60.0])), + ("gpt-5.4", (272_000, [5.0, 22.5, 5.0, 0.5])), + ("gpt-5.4-pro", (272_000, [60.0, 270.0, 60.0, 60.0])), + ("gemini-3.1-pro-preview", (200_000, [4.0, 18.0, 4.0, 0.4])), + ("gemini-2.5-pro", (200_000, [2.5, 15.0, 2.5, 0.25])), +]; + +const HAIKU_5_5_LONG_CONTEXT: (u32, [f64; 4]) = (100_000, [0.5, 2.5, 0.625, 0.05]); + +/// Returns pricing for a single request whose prompt (input plus cache reads and writes) +/// is `prompt_tokens` long, applying the long-context tiers that [`pricing_for_model`] +/// leaves out because it also prices usage aggregated across requests. +#[must_use] +pub fn pricing_for_request(model: &str, prompt_tokens: u32) -> Option { + let normalized = model.to_ascii_lowercase(); + let long_context = if family_version(&normalized, "haiku") == Some((5, 5)) { + Some(HAIKU_5_5_LONG_CONTEXT) + } else { + find_model(LONG_CONTEXT_PRICING, &normalized) + }; + match long_context { + Some((threshold, rates)) if prompt_tokens > threshold => { + Some(ModelPricing::from_rates(rates)) + } + _ => pricing_for_model(model), + } +} + fn openai_compatible_pricing(normalized: &str) -> Option { + find_model(OPENAI_COMPATIBLE_PRICING, normalized).map(ModelPricing::from_rates) +} + +fn find_model(table: &[(&str, T)], normalized: &str) -> Option { let id = normalized.rsplit_once('/').map_or(normalized, |(_, id)| id); - OPENAI_COMPATIBLE_PRICING + table .iter() .find(|(model, _)| { id.strip_prefix(model).is_some_and(|rest| { @@ -201,9 +268,7 @@ fn openai_compatible_pricing(normalized: &str) -> Option { }) }) }) - .map(|&(_, [input, output, cache_write, cache_read])| { - ModelPricing::per_million(input, output, cache_write, cache_read) - }) + .map(|&(_, value)| value) } /// Parses the `(major, minor)` version after `family` in ids such as @@ -354,7 +419,7 @@ impl UsageTracker { #[cfg(test)] mod tests { - use super::{format_usd, pricing_for_model, TokenUsage, UsageTracker}; + use super::{format_usd, pricing_for_model, pricing_for_request, TokenUsage, UsageTracker}; use crate::session::{ContentBlock, ConversationMessage, MessageRole, Session}; #[test] @@ -507,6 +572,13 @@ mod tests { assert_eq!(total_for("qwen/qwen-max"), "$1.1025"); assert_eq!(total_for("qwen-plus-latest"), "$0.2815"); assert_eq!(total_for("kimi-k2.5"), "$2.1943"); + assert_eq!(total_for("gpt-5.5-pro"), "$150.0000"); + assert_eq!(total_for("gpt-5.3-codex"), "$8.9250"); + assert_eq!(total_for("o3-pro-2025-06-10"), "$80.0000"); + assert_eq!(total_for("o3"), "$6.5000"); + assert_eq!(total_for("google/gemini-2.5-pro"), "$6.3750"); + assert_eq!(total_for("gemini-2.5-flash-lite"), "$0.3100"); + assert_eq!(total_for("gemini-3.1-pro-preview"), "$8.2000"); for unpriced in [ "o1-mini", "grok-3-mini", @@ -517,6 +589,31 @@ mod tests { } } + #[test] + fn applies_long_context_tiers_per_request() { + let rates = |model: &str, prompt_tokens: u32| { + let pricing = pricing_for_request(model, prompt_tokens).expect("known model pricing"); + ( + pricing.input_cost_per_million, + pricing.output_cost_per_million, + ) + }; + + assert_eq!(rates("claude-haiku-5-5", 100_000), (0.1, 0.5)); + assert_eq!(rates("claude-haiku-5-5-20260301", 100_001), (0.5, 2.5)); + assert_eq!(rates("claude-haiku-4-5", 500_000), (1.0, 5.0)); + assert_eq!(rates("gpt-5.4", 272_000), (2.5, 15.0)); + assert_eq!(rates("openai/gpt-5.4", 272_001), (5.0, 22.5)); + assert_eq!(rates("gpt-5.4-mini", 900_000), (0.75, 4.5)); + assert_eq!(rates("gemini-2.5-pro", 200_001), (2.5, 15.0)); + assert_eq!(rates("claude-sonnet-4-6", 900_000), (3.0, 15.0)); + assert_eq!( + pricing_for_request("gpt-5.5", 1_000), + pricing_for_model("gpt-5.5") + ); + assert_eq!(pricing_for_request("custom-model", 1_000_000), None); + } + #[test] fn marks_unknown_model_pricing_as_fallback() { let usage = TokenUsage {