Skip to content

Verify model pricing against provider pages and price long-context tiers #607

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
M3gA-Mind merged 4 commits into from
Nov 11, 2026
Merged

Verify model pricing against provider pages and price long-context tiers #607

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
4d62dff
Refresh the price table against provider pages and model long-context…
M3gA-Mind Nov 11, 2026
a2eb22b
Re-point renamed Mistral ids and drop retired models
M3gA-Mind Nov 11, 2026
a5438be
Drop mistral-medium from the price table
M3gA-Mind Nov 11, 2026
d7c0740
Price Mistral Large 4 under its own model ids
M3gA-Mind Nov 11, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
210 changes: 118 additions & 92 deletions crates/tinyagents-registry/src/pricing/data.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,17 +1,19 @@
//! July 2026 published price and context-window snapshot.
//! Published price and context-window snapshot, last checked 2026-11-11.

use super::ModelPrice;
use super::{LongContextTier, ModelPrice};

/// Published list prices and context windows for the default models the product
/// can route to.
///
/// Sources (captured [`super::PRICING_AS_OF`]): vendor pricing/model pages. Anthropic
/// price rows are authoritative (cached = 0.1× input, the documented cache-read
/// rate); other providers' cached rates use the published discount where known
/// and a conservative provider-typical fraction otherwise. Context windows are
/// the published maximums.
/// Each provider block names the page its rows were checked against and the
/// date ([`super::PRICING_AS_OF`]). Rates are standard (not batch) list prices;
/// the cached rate is the cache-read price. Models their provider retired are
/// removed rather than kept at stale prices. Context
/// windows are the published maximums. Rows with a long-context price step are
/// the short-context rates; [`LONG_CONTEXT_TIERS`] holds the step.
pub const KNOWN_MODEL_PRICING: &[ModelPrice] = &[
// ── Anthropic (authoritative prices; cache read = 0.1× input) ────────────
// ── Anthropic: platform.claude.com/docs/en/about-claude/pricing (2026-11-11).
// Cache read = 0.1× input; 4.6 and later bill the full 1M window flat.
ModelPrice {
provider: "anthropic",
model_id: "claude-fable-5",
Expand Down Expand Up @@ -76,36 +78,37 @@ pub const KNOWN_MODEL_PRICING: &[ModelPrice] = &[
output_per_mtok_usd: 5.00,
context_window: 200_000,
},
// ── OpenAI (cache read ≈ 0.25× input — published 75% off) ────────────────
// ── OpenAI: developers.openai.com/api/docs/pricing (2026-11-11).
// Cache read is 0.1× input on GPT-5.x and 0.25× on GPT-4.1 and o3.
ModelPrice {
provider: "openai",
model_id: "gpt-5.5",
input_per_mtok_usd: 5.00,
cached_input_per_mtok_usd: 1.25,
cached_input_per_mtok_usd: 0.50,
output_per_mtok_usd: 30.00,
context_window: 1_000_000,
context_window: 1_050_000,
},
ModelPrice {
provider: "openai",
model_id: "gpt-5.4",
input_per_mtok_usd: 2.50,
cached_input_per_mtok_usd: 0.625,
cached_input_per_mtok_usd: 0.25,
output_per_mtok_usd: 15.00,
context_window: 1_000_000,
context_window: 1_050_000,
},
ModelPrice {
provider: "openai",
model_id: "gpt-5.4-mini",
input_per_mtok_usd: 0.75,
cached_input_per_mtok_usd: 0.1875,
cached_input_per_mtok_usd: 0.075,
output_per_mtok_usd: 4.50,
context_window: 400_000,
},
ModelPrice {
provider: "openai",
model_id: "gpt-5.4-nano",
input_per_mtok_usd: 0.20,
cached_input_per_mtok_usd: 0.05,
cached_input_per_mtok_usd: 0.02,
output_per_mtok_usd: 1.25,
context_window: 400_000,
},
Expand Down Expand Up @@ -133,80 +136,51 @@ pub const KNOWN_MODEL_PRICING: &[ModelPrice] = &[
output_per_mtok_usd: 8.00,
context_window: 200_000,
},
// ── Google Gemini (cache read ≈ 0.25× input; 1M-token windows) ───────────
// ── Google: aistudio.google.com/docs/pricing (2026-11-11). Text cache read
// = 0.1× input. 2.5 Pro and Flash are deprecated and re-routed from
// 2026-12-01.
ModelPrice {
provider: "google",
model_id: "gemini-2.5-pro",
input_per_mtok_usd: 1.25,
cached_input_per_mtok_usd: 0.3125,
cached_input_per_mtok_usd: 0.125,
output_per_mtok_usd: 10.00,
context_window: 1_048_576,
},
ModelPrice {
provider: "google",
model_id: "gemini-2.5-flash",
input_per_mtok_usd: 0.30,
cached_input_per_mtok_usd: 0.075,
cached_input_per_mtok_usd: 0.03,
output_per_mtok_usd: 2.50,
context_window: 1_048_576,
},
ModelPrice {
provider: "google",
model_id: "gemini-2.5-flash-lite",
input_per_mtok_usd: 0.10,
cached_input_per_mtok_usd: 0.025,
cached_input_per_mtok_usd: 0.01,
output_per_mtok_usd: 0.40,
context_window: 1_048_576,
},
// ── DeepSeek (cache hit = 0.1× input, published) ─────────────────────────
ModelPrice {
provider: "deepseek",
model_id: "deepseek-chat",
input_per_mtok_usd: 0.14,
cached_input_per_mtok_usd: 0.014,
output_per_mtok_usd: 0.28,
context_window: 128_000,
},
ModelPrice {
provider: "deepseek",
model_id: "deepseek-reasoner",
input_per_mtok_usd: 0.55,
cached_input_per_mtok_usd: 0.055,
output_per_mtok_usd: 2.19,
context_window: 128_000,
},
// ── Moonshot Kimi (cache hit published) ──────────────────────────────────
// ── Moonshot: platform.kimi.ai/docs/pricing/chat (2026-11-11).
ModelPrice {
provider: "moonshot",
model_id: "kimi-k2.6",
input_per_mtok_usd: 0.95,
cached_input_per_mtok_usd: 0.16,
output_per_mtok_usd: 4.00,
context_window: 256_000,
},
ModelPrice {
provider: "moonshot",
model_id: "kimi-k2.5",
input_per_mtok_usd: 0.60,
cached_input_per_mtok_usd: 0.10,
output_per_mtok_usd: 3.00,
context_window: 256_000,
},
// ── Qwen / Alibaba (cache read ≈ 0.1× input) ─────────────────────────────
ModelPrice {
provider: "qwen",
model_id: "qwen3-max",
input_per_mtok_usd: 1.20,
cached_input_per_mtok_usd: 0.12,
output_per_mtok_usd: 6.00,
context_window: 256_000,
context_window: 262_144,
},
// ── Qwen: alibabacloud.com/help/en/model-studio/model-pricing, International
// (2026-11-11). Cache read is the explicit-cache rate (0.1× input);
// qwen-max has implicit cache only (0.2×).
ModelPrice {
provider: "qwen",
model_id: "qwen-max",
input_per_mtok_usd: 1.20,
cached_input_per_mtok_usd: 0.12,
output_per_mtok_usd: 6.00,
input_per_mtok_usd: 1.60,
cached_input_per_mtok_usd: 0.32,
output_per_mtok_usd: 6.40,
context_window: 256_000,
},
ModelPrice {
Expand All @@ -215,79 +189,131 @@ pub const KNOWN_MODEL_PRICING: &[ModelPrice] = &[
input_per_mtok_usd: 0.40,
cached_input_per_mtok_usd: 0.04,
output_per_mtok_usd: 1.20,
context_window: 256_000,
context_window: 1_000_000,
},
ModelPrice {
provider: "qwen",
model_id: "qwen-flash",
input_per_mtok_usd: 0.05,
cached_input_per_mtok_usd: 0.005,
output_per_mtok_usd: 0.40,
context_window: 256_000,
context_window: 1_000_000,
},
// ── Mistral (cache read ≈ 0.1× input) ────────────────────────────────────
// ── Mistral: docs.mistral.ai model cards (2026-11-11). mistral-large is
// Large 4 (26.10, published cache price), also listed under its own ids
// mistral-large-4 and mistral-large-4-0 (a one-digit suffix is not a
// snapshot decoration, so lookup would not reach the alias row); mistral-small is Small 4 (26.03);
// ministral-8b is Ministral 3 8B (25.12); codestral is 25.08. Small, 8B and
// Codestral publish no cache price, so theirs is an estimate (0.1× input).
ModelPrice {
provider: "mistral",
model_id: "mistral-large",
input_per_mtok_usd: 2.00,
cached_input_per_mtok_usd: 0.20,
output_per_mtok_usd: 6.00,
context_window: 128_000,
input_per_mtok_usd: 0.70,
cached_input_per_mtok_usd: 0.07,
output_per_mtok_usd: 2.10,
context_window: 1_000_000,
},
ModelPrice {
provider: "mistral",
model_id: "mistral-medium",
input_per_mtok_usd: 0.40,
cached_input_per_mtok_usd: 0.04,
output_per_mtok_usd: 2.00,
context_window: 128_000,
model_id: "mistral-large-4",
input_per_mtok_usd: 0.70,
cached_input_per_mtok_usd: 0.07,
output_per_mtok_usd: 2.10,
context_window: 1_000_000,
},
ModelPrice {
provider: "mistral",
model_id: "mistral-large-4-0",
input_per_mtok_usd: 0.70,
cached_input_per_mtok_usd: 0.07,
output_per_mtok_usd: 2.10,
context_window: 1_000_000,
},
ModelPrice {
provider: "mistral",
model_id: "mistral-small",
input_per_mtok_usd: 0.20,
cached_input_per_mtok_usd: 0.02,
input_per_mtok_usd: 0.15,
cached_input_per_mtok_usd: 0.015,
output_per_mtok_usd: 0.60,
context_window: 128_000,
context_window: 256_000,
},
ModelPrice {
provider: "mistral",
model_id: "codestral",
input_per_mtok_usd: 0.30,
cached_input_per_mtok_usd: 0.03,
output_per_mtok_usd: 0.90,
context_window: 256_000,
context_window: 128_000,
},
ModelPrice {
provider: "mistral",
model_id: "ministral-8b",
input_per_mtok_usd: 0.10,
cached_input_per_mtok_usd: 0.01,
output_per_mtok_usd: 0.10,
context_window: 128_000,
input_per_mtok_usd: 0.15,
cached_input_per_mtok_usd: 0.015,
output_per_mtok_usd: 0.15,
context_window: 256_000,
},
// ── MiniMax: platform.minimax.io/docs/guides/pricing (2026-11-11). M3's row
// is the ≤512K rate.
ModelPrice {
provider: "minimax",
model_id: "minimax-m3",
input_per_mtok_usd: 0.60,
cached_input_per_mtok_usd: 0.12,
output_per_mtok_usd: 2.40,
context_window: 1_000_000,
},
ModelPrice {
provider: "minimax",
model_id: "minimax-m2.7",
input_per_mtok_usd: 0.30,
cached_input_per_mtok_usd: 0.06,
output_per_mtok_usd: 1.20,
context_window: 204_800,
context_window: 1_000_000,
},
ModelPrice {
provider: "minimax",
model_id: "minimax-m2.7-highspeed",
];

/// Published long-context price steps, from the same pages as the rows. Above
/// `above_input_tokens` input tokens (cached tokens included) the whole call
/// bills at the step's rates.
pub const LONG_CONTEXT_TIERS: &[LongContextTier] = &[
// OpenAI: "Long context: >272K input tokens".
LongContextTier {
model_id: "gpt-5.5",
above_input_tokens: 272_000,
input_per_mtok_usd: 10.00,
cached_input_per_mtok_usd: 1.00,
output_per_mtok_usd: 45.00,
},
LongContextTier {
model_id: "gpt-5.4",
above_input_tokens: 272_000,
input_per_mtok_usd: 5.00,
cached_input_per_mtok_usd: 0.50,
output_per_mtok_usd: 22.50,
},
// Google: prompts >200K.
LongContextTier {
model_id: "gemini-2.5-pro",
above_input_tokens: 200_000,
input_per_mtok_usd: 2.50,
cached_input_per_mtok_usd: 0.25,
output_per_mtok_usd: 15.00,
},
// Qwen: the 256K-1M input tier (non-thinking), billed per request.
LongContextTier {
model_id: "qwen-plus",
above_input_tokens: 256_000,
input_per_mtok_usd: 1.20,
cached_input_per_mtok_usd: 0.12,
output_per_mtok_usd: 3.60,
},
LongContextTier {
model_id: "qwen-flash",
above_input_tokens: 256_000,
input_per_mtok_usd: 0.25,
cached_input_per_mtok_usd: 0.025,
output_per_mtok_usd: 2.00,
},
// MiniMax: "> 512k input tokens". The page does not say whether cached
// tokens count toward the boundary; they are counted here.
LongContextTier {
model_id: "minimax-m3",
above_input_tokens: 512_000,
input_per_mtok_usd: 0.60,
cached_input_per_mtok_usd: 0.06,
cached_input_per_mtok_usd: 0.12,
output_per_mtok_usd: 2.40,
context_window: 204_800,
},
];
Loading
Loading