Documentation
¶
Overview ¶
Package pricing provides a hardcoded pricing table for Claude models and cost/compaction helpers for the usage panel.
Prices are sourced from Anthropic's public pricing page and are hardcoded because Anthropic does not expose a machine-readable pricing API.
WARNING: These values may be stale. Verify against https://www.anthropic.com/pricing before making billing decisions. Last verified: 2026-07-02.
Index ¶
Constants ¶
This section is empty.
Variables ¶
var ( // Fable5 is claude-fable-5 — Anthropic's top tier, above Opus. // The 1M window is Fable's default and only size, at standard pricing. Fable5 = Model{ ID: "claude-fable-5", InputPerMillion: 10.00, OutputPerMillion: 50.00, CacheReadPerMillion: 1.00, CacheCreationPerMillion: 12.50, ContextLimit: 1_000_000, } // Opus48 is claude-opus-4-8 — current Opus generation. Opus48 = Model{ ID: "claude-opus-4-8", InputPerMillion: 5.00, OutputPerMillion: 25.00, CacheReadPerMillion: 0.50, CacheCreationPerMillion: 6.25, ContextLimit: 200_000, } // Opus47 is claude-opus-4-7 — previous Opus generation. // Opus dropped to $5/$25 with the 4.5 release. Opus47 = Model{ ID: "claude-opus-4-7", InputPerMillion: 5.00, OutputPerMillion: 25.00, CacheReadPerMillion: 0.50, CacheCreationPerMillion: 6.25, ContextLimit: 200_000, } // Sonnet5 is claude-sonnet-5 — current Sonnet generation. // Sticker price ($3/$15); the 2026 introductory discount ($2/$10 // through 2026-08-31) is intentionally not modeled. Sonnet5 = Model{ ID: "claude-sonnet-5", InputPerMillion: 3.00, OutputPerMillion: 15.00, CacheReadPerMillion: 0.30, CacheCreationPerMillion: 3.75, ContextLimit: 200_000, } // Sonnet46 is claude-sonnet-4-6 — previous Sonnet generation. Sonnet46 = Model{ ID: "claude-sonnet-4-6", InputPerMillion: 3.00, OutputPerMillion: 15.00, CacheReadPerMillion: 0.30, CacheCreationPerMillion: 3.75, ContextLimit: 200_000, } // Sonnet45 is claude-sonnet-4-5 — older Sonnet generation. Its 1M // window was a long-context beta billed at a premium (see OneMPremium). Sonnet45 = Model{ ID: "claude-sonnet-4-5", InputPerMillion: 3.00, OutputPerMillion: 15.00, CacheReadPerMillion: 0.30, CacheCreationPerMillion: 3.75, ContextLimit: 200_000, OneMPremium: true, } // Haiku45 is claude-haiku-4-5 — fastest, lowest cost. Haiku45 = Model{ ID: "claude-haiku-4-5", InputPerMillion: 1.00, OutputPerMillion: 5.00, CacheReadPerMillion: 0.10, CacheCreationPerMillion: 1.25, ContextLimit: 200_000, } // Opus45 is claude-opus-4-5 — the release that cut Opus to $5/$25. Opus45 = Model{ ID: "claude-opus-4-5", InputPerMillion: 5.00, OutputPerMillion: 25.00, CacheReadPerMillion: 0.50, CacheCreationPerMillion: 6.25, ContextLimit: 200_000, } )
Predefined models. Prices in USD/million tokens; context in tokens. Sources: https://www.anthropic.com/pricing (verified 2026-07-02). Cache rates follow Anthropic's standard multipliers: reads = 0.1× input, 5-minute writes = 1.25× input. Stale warning: update when Anthropic publishes new prices.
Functions ¶
func CompactionPercent ¶
CompactionPercent returns a value in [0, 1] representing how full the model's context window is for the given single-turn usage snapshot.
u should be the LATEST assistant event's usage (not a running sum across all turns). The context fill is: input + cache_read + cache_creation — all tokens sent to the model in that one API call. Returns 0 when m.ContextLimit == 0 to avoid divide-by-zero.
func ContextTokens ¶
ContextTokens returns the full context size of a single API call: input + cache_read + cache_creation.
This is the correct value to use for compaction percentage calculation because it represents the total number of tokens that were sent to the model in a single turn (cached or not). Use this with the LATEST assistant event's usage, not a running sum — the running sum inflates beyond any meaningful context limit.
func Cost ¶
Cost computes the total USD cost for the given usage on the given model. Returns 0 when m.ContextLimit == 0 (zero model) to avoid divide-by-zero.
func TotalTokens ¶
TotalTokens returns the "billed activity" token count for display: input + output + cache_creation.
cache_read_input_tokens is intentionally excluded from this total because it is a per-turn value that records how many cached tokens were read on each individual API call. The same cached content is read again on every turn that uses it, so summing cache_read across turns counts the same tokens many times (e.g. a 5h session can accumulate 65M "cache read" tokens while the actual context is only ~750k). Including cache_read in the display total produces impossibly high numbers.
Cost() still applies the correct 0.1× rate to cache_read tokens for billing purposes — that is unaffected by this change.
Types ¶
type Model ¶
type Model struct {
// ID is the canonical model identifier as it appears in the JSONL
// "message.model" field, e.g. "claude-opus-4-7".
ID string
// InputPerMillion is the cost in USD per one million input tokens.
InputPerMillion float64
// OutputPerMillion is the cost in USD per one million output tokens.
OutputPerMillion float64
// CacheReadPerMillion is the cost in USD per one million cache-read tokens.
CacheReadPerMillion float64
// CacheCreationPerMillion is the cost in USD per one million
// cache-creation tokens.
CacheCreationPerMillion float64
// ContextLimit is the maximum context window in tokens for this model.
// Used to compute CompactionPercent.
//
// Note: this reflects the window Claude Code runs the model with by
// default (200K for most models), not the API maximum. Sessions running
// under the 1M window are identified by the "[1m]" model-ID suffix and
// resolved via Lookup, which overrides ContextLimit to 1M.
ContextLimit int64
// OneMPremium is true for legacy models whose 1M long-context beta was
// billed at a premium (2× input / 1.5× output above the 200K threshold).
// Models from the 4.6 era onward serve the 1M window at standard
// per-token pricing, so their [1m] variant only changes ContextLimit.
OneMPremium bool
}
Model holds the pricing and context metadata for a Claude model.
func Lookup ¶
Lookup returns the Model for the given model ID. If the ID is not found in the pricing table, it returns a generic fallback model with mid-tier pricing rather than a zero-cost model.
1M context variants: when id ends with "[1m]", ContextLimit is overridden to 1_000_000. Models from the 4.6 era onward serve the 1M window at standard per-token pricing (no long-context premium), so prices stay at the base rate; only legacy OneMPremium models (Sonnet 4.5's long-context beta) apply the 2× input / 1.5× output premium, approximated flat since per-request tiering isn't visible in the JSONL.
Source Files