@alexeiled/claude-router 1.20.1 → 1.21.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/lib/policy.mjs CHANGED
@@ -130,7 +130,7 @@ export function cashGate(config, route, facts, now, costs, routeFor = (tier) =>
130
130
  if (routeSpec(config, route).billing !== 'credits') return { blocked: false };
131
131
  if (costs.isWarm(facts.models[costs.routeCacheKey(config, route, facts.effort)], now, config.cache))
132
132
  return { blocked: false };
133
- const cold = costs.coldWriteUsd(config, route.model, costs.nextContextTokens(facts));
133
+ const cold = costs.coldWriteUsd(config, route.model, costs.nextContextTokens(facts), now);
134
134
  if (cold <= config.policy.cashCapUsd) return { blocked: false };
135
135
  const fallback = [...TIERS].reverse().find((t) => routeSpec(config, routeFor(t)).billing === 'plan');
136
136
  const estimate = {
package/lib/prices.mjs ADDED
@@ -0,0 +1,113 @@
1
+ // Generated by scripts/check-prices.mjs from each vendor's official price page: `npm run prices:check -- --write`
2
+ // rewrites this file, and the weekly price check proposes the change as a pull request. USD per million tokens.
3
+ // `checked` is the date the prices last changed or a person confirmed them against the page. `offPeak` multiplies
4
+ // every rate outside the peak hours (UTC days 0-6 from Sunday, [from, to) hours). Limit: DeepSeek also bills Chinese
5
+ // public holidays off-peak, which this does not know, so a holiday prices at peak, the higher bound.
6
+ export const PRICE_SOURCES = {
7
+ anthropic: {
8
+ url: 'https://platform.claude.com/docs/en/about-claude/pricing.md',
9
+ checked: '2026-11-04',
10
+ },
11
+ deepseek: {
12
+ url: 'https://api-docs.deepseek.com/quick_start/pricing.md',
13
+ checked: '2026-11-04',
14
+ },
15
+ kimi: {
16
+ url: 'https://platform.kimi.ai/docs/pricing/chat',
17
+ checked: '2026-11-04',
18
+ },
19
+ minimax: {
20
+ url: 'https://platform.minimax.io/docs/guides/pricing-paygo.md',
21
+ checked: '2026-11-04',
22
+ },
23
+ zai: {
24
+ url: 'https://docs.z.ai/guides/overview/pricing.md',
25
+ checked: '2026-11-04',
26
+ },
27
+ mimo: {
28
+ url: 'https://mimo.mi.com/docs/en-US/price/pay-as-you-go',
29
+ checked: '2026-11-04',
30
+ manual: true,
31
+ },
32
+ };
33
+
34
+ export const PRICES = {
35
+ 'claude-opus-5-5': {
36
+ source: 'anthropic',
37
+ input: 4,
38
+ output: 20,
39
+ cacheRead: 0.2,
40
+ },
41
+ 'claude-sonnet-5-5': {
42
+ source: 'anthropic',
43
+ input: 2,
44
+ output: 10,
45
+ cacheRead: 0.1,
46
+ },
47
+ 'claude-haiku-5-5': {
48
+ source: 'anthropic',
49
+ input: 0.1,
50
+ output: 0.5,
51
+ cacheRead: 0.01,
52
+ longContext: {
53
+ above: 100000,
54
+ multiplier: 5,
55
+ },
56
+ },
57
+ 'deepseek-flash[1m]': {
58
+ source: 'deepseek',
59
+ input: 0.3,
60
+ output: 1.2,
61
+ cacheRead: 0.006,
62
+ offPeak: {
63
+ multiplier: 0.5,
64
+ peakDays: [1, 2, 3, 4, 5],
65
+ peakHoursUtc: [
66
+ [1, 4],
67
+ [6, 10],
68
+ ],
69
+ },
70
+ },
71
+ 'kimi-k3.1-pro[1m]': {
72
+ source: 'kimi',
73
+ input: 2,
74
+ output: 10,
75
+ cacheRead: 0.2,
76
+ },
77
+ 'MiniMax-M3.1-Flash[1m]': {
78
+ source: 'minimax',
79
+ input: 0.15,
80
+ output: 0.6,
81
+ cacheRead: 0.015,
82
+ },
83
+ 'glm-5.3-flash[1m]': {
84
+ source: 'zai',
85
+ input: 0.15,
86
+ output: 0.5,
87
+ cacheRead: 0.03,
88
+ },
89
+ 'glm-5.3[1m]': {
90
+ source: 'zai',
91
+ input: 1.4,
92
+ output: 4.4,
93
+ cacheRead: 0.26,
94
+ },
95
+ 'mimo-v2.6-flash[1m]': {
96
+ source: 'mimo',
97
+ input: 0.14,
98
+ output: 0.28,
99
+ cacheRead: 0.0028,
100
+ },
101
+ 'mimo-v2.6-pro[1m]': {
102
+ source: 'mimo',
103
+ input: 0.435,
104
+ output: 0.87,
105
+ cacheRead: 0.0036,
106
+ },
107
+ };
108
+
109
+ // A model's rates for a router.json entry, without the bookkeeping field.
110
+ export function listPrice(id) {
111
+ const { source, ...rates } = PRICES[id];
112
+ return rates;
113
+ }
package/lib/providers.mjs CHANGED
@@ -1,3 +1,5 @@
1
+ import { listPrice } from './prices.mjs';
2
+
1
3
  // Execution presets, not classifier services. Prices are API list-price estimates, never a plan bill.
2
4
  // Sources and the limits of these starting routes are in docs/configuration.md#provider-profiles.
3
5
  const EFFORTS = ['low', 'high', 'max'];
@@ -22,9 +24,7 @@ export const PROVIDER_DEFAULTS = {
22
24
  models: {
23
25
  flash: {
24
26
  id: 'MiniMax-M3.1-Flash[1m]',
25
- input: 0.15,
26
- output: 0.6,
27
- cacheRead: 0.015,
27
+ ...listPrice('MiniMax-M3.1-Flash[1m]'),
28
28
  contextWindow: 1_000_000,
29
29
  billing: 'credits',
30
30
  // xhigh is documented but some Claude Code versions serialize it as high; use max instead.
@@ -49,9 +49,7 @@ export const PROVIDER_DEFAULTS = {
49
49
  models: {
50
50
  flash: {
51
51
  id: 'mimo-v2.6-flash[1m]',
52
- input: 0.14,
53
- output: 0.28,
54
- cacheRead: 0.0028,
52
+ ...listPrice('mimo-v2.6-flash[1m]'),
55
53
  contextWindow: 1_000_000,
56
54
  billing: 'credits',
57
55
  // The Anthropic endpoint documents binary thinking, not granular effort. Keep its default thinking.
@@ -61,9 +59,7 @@ export const PROVIDER_DEFAULTS = {
61
59
  },
62
60
  pro: {
63
61
  id: 'mimo-v2.6-pro[1m]',
64
- input: 0.435,
65
- output: 0.87,
66
- cacheRead: 0.0036,
62
+ ...listPrice('mimo-v2.6-pro[1m]'),
67
63
  contextWindow: 1_000_000,
68
64
  billing: 'credits',
69
65
  efforts: [],
@@ -86,9 +82,7 @@ export const PROVIDER_DEFAULTS = {
86
82
  models: {
87
83
  k3: {
88
84
  id: 'kimi-k3.1-pro[1m]',
89
- input: 2,
90
- output: 10,
91
- cacheRead: 0.2,
85
+ ...listPrice('kimi-k3.1-pro[1m]'),
92
86
  contextWindow: 1_048_576,
93
87
  billing: 'credits',
94
88
  efforts: EFFORTS,
@@ -110,9 +104,7 @@ export const PROVIDER_DEFAULTS = {
110
104
  models: {
111
105
  flash: {
112
106
  id: 'deepseek-flash[1m]',
113
- input: 0.3,
114
- output: 1.2,
115
- cacheRead: 0.006,
107
+ ...listPrice('deepseek-flash[1m]'),
116
108
  contextWindow: 1_000_000,
117
109
  billing: 'credits',
118
110
  efforts: EFFORTS,
@@ -136,9 +128,7 @@ export const PROVIDER_DEFAULTS = {
136
128
  models: {
137
129
  flash: {
138
130
  id: 'glm-5.3-flash[1m]',
139
- input: 0.15,
140
- output: 0.5,
141
- cacheRead: 0.03,
131
+ ...listPrice('glm-5.3-flash[1m]'),
142
132
  contextWindow: 1_000_000,
143
133
  billing: 'credits',
144
134
  efforts: EFFORTS,
@@ -147,9 +137,7 @@ export const PROVIDER_DEFAULTS = {
147
137
  },
148
138
  glm: {
149
139
  id: 'glm-5.3[1m]',
150
- input: 1.4,
151
- output: 4.4,
152
- cacheRead: 0.26,
140
+ ...listPrice('glm-5.3[1m]'),
153
141
  contextWindow: 1_000_000,
154
142
  billing: 'credits',
155
143
  efforts: EFFORTS,
package/lib/route.mjs CHANGED
@@ -67,7 +67,7 @@ export function emptyLoop(config, model) {
67
67
  // The message count at the last review applied inside a turn: tool calls before it were already judged, so only
68
68
  // later ones can move the route up again (observedMove).
69
69
  observedFrom: 0,
70
- // Each configured model's simulated cache, by models key, for the routing-vs-your-model readout (compareModels in
70
+ // Each configured model's simulated cache, by models key, for the routing-vs-no-routing readout (compareModels in
71
71
  // savings.mjs).
72
72
  yoursBy: {},
73
73
  };
package/lib/savings.mjs CHANGED
@@ -38,9 +38,15 @@ export function modelAlias(config, id) {
38
38
  }
39
39
 
40
40
  // One request's list price in USD: `counts` are its uncached input, cache read, cache write and output tokens, the
41
- // writes at `writeMultiplier` times input. Long-context rates follow the whole prompt.
42
- export function requestUsd(model, { input, cacheRead, cacheWrite, output }, writeMultiplier, cacheMode = 'explicit') {
43
- const rates = ratesAt(model, input + cacheRead + cacheWrite);
41
+ // writes at `writeMultiplier` times input. Long-context rates follow the whole prompt, off-peak rates the time `at`.
42
+ export function requestUsd(
43
+ model,
44
+ { input, cacheRead, cacheWrite, output },
45
+ writeMultiplier,
46
+ cacheMode = 'explicit',
47
+ at,
48
+ ) {
49
+ const rates = ratesAt(model, input + cacheRead + cacheWrite, at);
44
50
  const inputUsd =
45
51
  cacheMode === 'disabled'
46
52
  ? rates.input * (input + cacheRead + cacheWrite)
@@ -49,7 +55,7 @@ export function requestUsd(model, { input, cacheRead, cacheWrite, output }, writ
49
55
  }
50
56
 
51
57
  // Actual observed counters can be priced without inventing the cache on another route.
52
- export function observedReplyUsd(config, usage, ttl) {
58
+ export function observedReplyUsd(config, usage, ttl, at) {
53
59
  const counts = {
54
60
  input: usage?.input_tokens,
55
61
  cacheRead: usage?.cache_read_input_tokens,
@@ -65,7 +71,7 @@ export function observedReplyUsd(config, usage, ttl) {
65
71
  !Object.values(counts).every((n) => Number.isFinite(n) && n >= 0)
66
72
  )
67
73
  return null;
68
- return requestUsd(model, counts, write, config.cache.mode);
74
+ return requestUsd(model, counts, write, config.cache.mode, at);
69
75
  }
70
76
 
71
77
  const truthy = (value) => /^(1|true|yes|on)$/i.test(String(value ?? '').trim());
@@ -138,7 +144,7 @@ export function compareReply(config, state, { usage, served, yours, ttl, now })
138
144
  return { state: next, reply: null };
139
145
  // Implicit caches have no guaranteed TTL or cross-effort hit rate. Only identical observed routes compare exactly.
140
146
  if (config.cache.mode === 'automatic' && !(onYours && (prior?.onYours ?? true))) return { state: next, reply: null };
141
- const routedUsd = requestUsd(servedSpec, counts, write, config.cache.mode);
147
+ const routedUsd = requestUsd(servedSpec, counts, write, config.cache.mode, now);
142
148
  let yoursUsd = routedUsd;
143
149
  let servedOnYoursUsd = routedUsd;
144
150
  if (!(onYours && (prior?.onYours ?? true))) {
@@ -150,8 +156,8 @@ export function compareReply(config, state, { usage, served, yours, ttl, now })
150
156
  prior.total > 0;
151
157
  const cacheRead = Math.max(counts.cacheRead, warm ? Math.min(prior.total, total - counts.input) : 0);
152
158
  const single = { ...counts, cacheRead, cacheWrite: total - counts.input - cacheRead };
153
- yoursUsd = requestUsd(yoursSpec, single, write, config.cache.mode);
154
- servedOnYoursUsd = requestUsd(servedSpec, single, write, config.cache.mode);
159
+ yoursUsd = requestUsd(yoursSpec, single, write, config.cache.mode, now);
160
+ servedOnYoursUsd = requestUsd(servedSpec, single, write, config.cache.mode, now);
155
161
  }
156
162
  const modelUsd = servedOnYoursUsd - yoursUsd;
157
163
  return {
package/lib/subagents.mjs CHANGED
@@ -172,7 +172,7 @@ export function spawnRecord({ sessionId, spawn, pass, choice, routedModel, coreM
172
172
  const isCount = (n) => Number.isFinite(n) && n >= 0;
173
173
 
174
174
  // The record with one reply added. A reply is priced only when both its model and the comparison's have a price.
175
- export function addAgentReply(config, agent, usage, ttl) {
175
+ export function addAgentReply(config, agent, usage, ttl, at) {
176
176
  const counts = {
177
177
  input: usage?.input_tokens,
178
178
  cacheRead: usage?.cache_read_input_tokens,
@@ -188,8 +188,8 @@ export function addAgentReply(config, agent, usage, ttl) {
188
188
  if (!agent.pass && !priced(other)) return next;
189
189
  if (config.cache.mode === 'automatic' && other && actual !== other) return next;
190
190
  next.pricedRequests += 1;
191
- next.actualUsd += requestUsd(actual, counts, write, config.cache.mode);
192
- if (other) next.otherUsd += requestUsd(other, counts, write, config.cache.mode);
191
+ next.actualUsd += requestUsd(actual, counts, write, config.cache.mode, at);
192
+ if (other) next.otherUsd += requestUsd(other, counts, write, config.cache.mode, at);
193
193
  return next;
194
194
  }
195
195
 
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@alexeiled/claude-router",
3
- "version": "1.20.1",
3
+ "version": "1.21.0",
4
4
  "description": "Native Claude Code model and effort routing with Anthropic, DeepSeek, z.ai, MiniMax, MiMo and Kimi defaults, custom provider profiles, and classifier advice.",
5
5
  "license": "MIT",
6
6
  "author": "Alexei Ledenev",
@@ -37,6 +37,7 @@
37
37
  "test:providers": "node experiments/mod-router/scripts/provider-acceptance.mjs",
38
38
  "format": "tools/node_modules/.bin/biome format --write .",
39
39
  "docs:images": "node scripts/render-doc-images.mjs",
40
+ "prices:check": "node scripts/check-prices.mjs",
40
41
  "docs:modes": "node scripts/mode-matrix.mjs",
41
42
  "probe:results": "node scripts/probe-results.mjs",
42
43
  "pack:dry": "npm pack --dry-run",