sidekick-shared 0.28.1 → 0.28.2
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/CHANGELOG.md +6 -0
- package/dist/esm/modelContext.js +32 -0
- package/dist/esm/modelInfo.js +186 -0
- package/dist/modelContext.js +32 -0
- package/dist/modelInfo.js +186 -0
- package/package.json +1 -1
package/CHANGELOG.md
CHANGED
|
@@ -5,6 +5,12 @@ All notable changes to sidekick-shared will be documented in this file.
|
|
|
5
5
|
The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
|
|
6
6
|
and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
|
|
7
7
|
|
|
8
|
+
## [0.28.2] - 2026-10-30
|
|
9
|
+
|
|
10
|
+
### Fixed
|
|
11
|
+
|
|
12
|
+
- Add verified offline pricing and context baselines for GPT-6.1 Sol/Terra, Claude Sonnet/Haiku 5.5 (dashed and dotted), Gemini 4 Argon/3.8 Flash, DeepSeek Flash/V4 Pro (including accepted Flash aliases), and GLM-5.3/Flash/FlashX. Direct-provider IDs resolve offline too. Preserve runtime overrides and observed account limits; document standard/short-context rates, DeepSeek peak rates, and Gemini promotions. Fixes #18.
|
|
13
|
+
|
|
8
14
|
## [0.28.1] - 2026-10-18
|
|
9
15
|
|
|
10
16
|
### Fixed
|
package/dist/esm/modelContext.js
CHANGED
|
@@ -30,6 +30,38 @@
|
|
|
30
30
|
import { resolveModelAlias } from './modelAliases';
|
|
31
31
|
/** Known model context window sizes (in tokens); current models verified 2026-09-05. */
|
|
32
32
|
const MODEL_CONTEXT_SIZES = {
|
|
33
|
+
// Published API limits verified 2026-10-30; GLM's API documents 1M even
|
|
34
|
+
// though local weights may support more. Catalog/observed overrides win.
|
|
35
|
+
'gpt-6.1-sol': 1_050_000,
|
|
36
|
+
'openai/gpt-6.1-sol': 1_050_000,
|
|
37
|
+
'gpt-6.1-terra': 1_050_000,
|
|
38
|
+
'openai/gpt-6.1-terra': 1_050_000,
|
|
39
|
+
'claude-sonnet-5-5': 1_000_000,
|
|
40
|
+
'anthropic/claude-sonnet-5-5': 1_000_000,
|
|
41
|
+
'claude-sonnet-5.5': 1_000_000,
|
|
42
|
+
'anthropic/claude-sonnet-5.5': 1_000_000,
|
|
43
|
+
'claude-haiku-5-5': 1_000_000,
|
|
44
|
+
'anthropic/claude-haiku-5-5': 1_000_000,
|
|
45
|
+
'claude-haiku-5.5': 1_000_000,
|
|
46
|
+
'anthropic/claude-haiku-5.5': 1_000_000,
|
|
47
|
+
'gemini-4-argon': 1_048_576,
|
|
48
|
+
'gemini/gemini-4-argon': 1_048_576,
|
|
49
|
+
'gemini-3.8-flash': 1_048_576,
|
|
50
|
+
'gemini/gemini-3.8-flash': 1_048_576,
|
|
51
|
+
'deepseek-flash': 1_000_000,
|
|
52
|
+
'deepseek/deepseek-flash': 1_000_000,
|
|
53
|
+
'deepseek-v4-flash': 1_000_000,
|
|
54
|
+
'deepseek/deepseek-v4-flash': 1_000_000,
|
|
55
|
+
'deepseek-v4-flash-vision-exp': 1_000_000,
|
|
56
|
+
'deepseek/deepseek-v4-flash-vision-exp': 1_000_000,
|
|
57
|
+
'deepseek-v4-pro': 1_000_000,
|
|
58
|
+
'deepseek/deepseek-v4-pro': 1_000_000,
|
|
59
|
+
'glm-5.3': 1_000_000,
|
|
60
|
+
'zai/glm-5.3': 1_000_000,
|
|
61
|
+
'glm-5.3-flash': 1_000_000,
|
|
62
|
+
'zai/glm-5.3-flash': 1_000_000,
|
|
63
|
+
'glm-5.3-flashx': 1_000_000,
|
|
64
|
+
'zai/glm-5.3-flashx': 1_000_000,
|
|
33
65
|
// Claude — native 1M context (Fable 5+, Opus 4.6+, Sonnet 4.6+)
|
|
34
66
|
'claude-fable-5-1': 1_000_000,
|
|
35
67
|
'claude-fable-5.1': 1_000_000,
|
package/dist/esm/modelInfo.js
CHANGED
|
@@ -37,6 +37,192 @@ export { MODEL_PROVIDER_IDS } from './providerIds';
|
|
|
37
37
|
* prefix matching cannot bridge the two, so both are needed.
|
|
38
38
|
*/
|
|
39
39
|
const PRICING_TABLE = {
|
|
40
|
+
// Current coding models verified 2026-10-30. Standard short-context rates;
|
|
41
|
+
// DeepSeek uses peak rates. Gemini promotions end 2026-11-30 (Argon) and
|
|
42
|
+
// 2026-12-31 (Flash). Haiku 5.5 costs more above 100K prompt tokens.
|
|
43
|
+
// Sources and estimation limits: docs/configuration/model-resolution.md.
|
|
44
|
+
// Explicit direct-provider keys support offline OpenCode model IDs without
|
|
45
|
+
// stripping arbitrary deployment prefixes or changing alias resolution.
|
|
46
|
+
'gpt-6.1-sol': {
|
|
47
|
+
inputCostPerMillion: 2,
|
|
48
|
+
outputCostPerMillion: 10,
|
|
49
|
+
cacheWriteCostPerMillion: 2.5,
|
|
50
|
+
cacheReadCostPerMillion: 0.1,
|
|
51
|
+
},
|
|
52
|
+
'openai/gpt-6.1-sol': {
|
|
53
|
+
inputCostPerMillion: 2,
|
|
54
|
+
outputCostPerMillion: 10,
|
|
55
|
+
cacheWriteCostPerMillion: 2.5,
|
|
56
|
+
cacheReadCostPerMillion: 0.1,
|
|
57
|
+
},
|
|
58
|
+
'gpt-6.1-terra': {
|
|
59
|
+
inputCostPerMillion: 1,
|
|
60
|
+
outputCostPerMillion: 5,
|
|
61
|
+
cacheWriteCostPerMillion: 1.25,
|
|
62
|
+
cacheReadCostPerMillion: 0.05,
|
|
63
|
+
},
|
|
64
|
+
'openai/gpt-6.1-terra': {
|
|
65
|
+
inputCostPerMillion: 1,
|
|
66
|
+
outputCostPerMillion: 5,
|
|
67
|
+
cacheWriteCostPerMillion: 1.25,
|
|
68
|
+
cacheReadCostPerMillion: 0.05,
|
|
69
|
+
},
|
|
70
|
+
'claude-sonnet-5-5': {
|
|
71
|
+
inputCostPerMillion: 2,
|
|
72
|
+
outputCostPerMillion: 10,
|
|
73
|
+
cacheWriteCostPerMillion: 2.5,
|
|
74
|
+
cacheReadCostPerMillion: 0.1,
|
|
75
|
+
},
|
|
76
|
+
'anthropic/claude-sonnet-5-5': {
|
|
77
|
+
inputCostPerMillion: 2,
|
|
78
|
+
outputCostPerMillion: 10,
|
|
79
|
+
cacheWriteCostPerMillion: 2.5,
|
|
80
|
+
cacheReadCostPerMillion: 0.1,
|
|
81
|
+
},
|
|
82
|
+
'claude-sonnet-5.5': {
|
|
83
|
+
inputCostPerMillion: 2,
|
|
84
|
+
outputCostPerMillion: 10,
|
|
85
|
+
cacheWriteCostPerMillion: 2.5,
|
|
86
|
+
cacheReadCostPerMillion: 0.1,
|
|
87
|
+
},
|
|
88
|
+
'anthropic/claude-sonnet-5.5': {
|
|
89
|
+
inputCostPerMillion: 2,
|
|
90
|
+
outputCostPerMillion: 10,
|
|
91
|
+
cacheWriteCostPerMillion: 2.5,
|
|
92
|
+
cacheReadCostPerMillion: 0.1,
|
|
93
|
+
},
|
|
94
|
+
'claude-haiku-5-5': {
|
|
95
|
+
inputCostPerMillion: 0.1,
|
|
96
|
+
outputCostPerMillion: 0.5,
|
|
97
|
+
cacheWriteCostPerMillion: 0.125,
|
|
98
|
+
cacheReadCostPerMillion: 0.01,
|
|
99
|
+
},
|
|
100
|
+
'anthropic/claude-haiku-5-5': {
|
|
101
|
+
inputCostPerMillion: 0.1,
|
|
102
|
+
outputCostPerMillion: 0.5,
|
|
103
|
+
cacheWriteCostPerMillion: 0.125,
|
|
104
|
+
cacheReadCostPerMillion: 0.01,
|
|
105
|
+
},
|
|
106
|
+
'claude-haiku-5.5': {
|
|
107
|
+
inputCostPerMillion: 0.1,
|
|
108
|
+
outputCostPerMillion: 0.5,
|
|
109
|
+
cacheWriteCostPerMillion: 0.125,
|
|
110
|
+
cacheReadCostPerMillion: 0.01,
|
|
111
|
+
},
|
|
112
|
+
'anthropic/claude-haiku-5.5': {
|
|
113
|
+
inputCostPerMillion: 0.1,
|
|
114
|
+
outputCostPerMillion: 0.5,
|
|
115
|
+
cacheWriteCostPerMillion: 0.125,
|
|
116
|
+
cacheReadCostPerMillion: 0.01,
|
|
117
|
+
},
|
|
118
|
+
'gemini-4-argon': {
|
|
119
|
+
inputCostPerMillion: 2,
|
|
120
|
+
outputCostPerMillion: 10,
|
|
121
|
+
cacheWriteCostPerMillion: 0,
|
|
122
|
+
cacheReadCostPerMillion: 0.1,
|
|
123
|
+
},
|
|
124
|
+
'gemini/gemini-4-argon': {
|
|
125
|
+
inputCostPerMillion: 2,
|
|
126
|
+
outputCostPerMillion: 10,
|
|
127
|
+
cacheWriteCostPerMillion: 0,
|
|
128
|
+
cacheReadCostPerMillion: 0.1,
|
|
129
|
+
},
|
|
130
|
+
'gemini-3.8-flash': {
|
|
131
|
+
inputCostPerMillion: 0.75,
|
|
132
|
+
outputCostPerMillion: 3.75,
|
|
133
|
+
cacheWriteCostPerMillion: 0,
|
|
134
|
+
cacheReadCostPerMillion: 0.075,
|
|
135
|
+
},
|
|
136
|
+
'gemini/gemini-3.8-flash': {
|
|
137
|
+
inputCostPerMillion: 0.75,
|
|
138
|
+
outputCostPerMillion: 3.75,
|
|
139
|
+
cacheWriteCostPerMillion: 0,
|
|
140
|
+
cacheReadCostPerMillion: 0.075,
|
|
141
|
+
},
|
|
142
|
+
'deepseek-flash': {
|
|
143
|
+
inputCostPerMillion: 0.3,
|
|
144
|
+
outputCostPerMillion: 1.2,
|
|
145
|
+
cacheWriteCostPerMillion: 0,
|
|
146
|
+
cacheReadCostPerMillion: 0.006,
|
|
147
|
+
},
|
|
148
|
+
'deepseek/deepseek-flash': {
|
|
149
|
+
inputCostPerMillion: 0.3,
|
|
150
|
+
outputCostPerMillion: 1.2,
|
|
151
|
+
cacheWriteCostPerMillion: 0,
|
|
152
|
+
cacheReadCostPerMillion: 0.006,
|
|
153
|
+
},
|
|
154
|
+
'deepseek-v4-flash': {
|
|
155
|
+
inputCostPerMillion: 0.3,
|
|
156
|
+
outputCostPerMillion: 1.2,
|
|
157
|
+
cacheWriteCostPerMillion: 0,
|
|
158
|
+
cacheReadCostPerMillion: 0.006,
|
|
159
|
+
},
|
|
160
|
+
'deepseek/deepseek-v4-flash': {
|
|
161
|
+
inputCostPerMillion: 0.3,
|
|
162
|
+
outputCostPerMillion: 1.2,
|
|
163
|
+
cacheWriteCostPerMillion: 0,
|
|
164
|
+
cacheReadCostPerMillion: 0.006,
|
|
165
|
+
},
|
|
166
|
+
'deepseek-v4-flash-vision-exp': {
|
|
167
|
+
inputCostPerMillion: 0.3,
|
|
168
|
+
outputCostPerMillion: 1.2,
|
|
169
|
+
cacheWriteCostPerMillion: 0,
|
|
170
|
+
cacheReadCostPerMillion: 0.006,
|
|
171
|
+
},
|
|
172
|
+
'deepseek/deepseek-v4-flash-vision-exp': {
|
|
173
|
+
inputCostPerMillion: 0.3,
|
|
174
|
+
outputCostPerMillion: 1.2,
|
|
175
|
+
cacheWriteCostPerMillion: 0,
|
|
176
|
+
cacheReadCostPerMillion: 0.006,
|
|
177
|
+
},
|
|
178
|
+
'deepseek-v4-pro': {
|
|
179
|
+
inputCostPerMillion: 1.32,
|
|
180
|
+
outputCostPerMillion: 3.96,
|
|
181
|
+
cacheWriteCostPerMillion: 0,
|
|
182
|
+
cacheReadCostPerMillion: 0.044,
|
|
183
|
+
},
|
|
184
|
+
'deepseek/deepseek-v4-pro': {
|
|
185
|
+
inputCostPerMillion: 1.32,
|
|
186
|
+
outputCostPerMillion: 3.96,
|
|
187
|
+
cacheWriteCostPerMillion: 0,
|
|
188
|
+
cacheReadCostPerMillion: 0.044,
|
|
189
|
+
},
|
|
190
|
+
'glm-5.3': {
|
|
191
|
+
inputCostPerMillion: 1.4,
|
|
192
|
+
outputCostPerMillion: 4.4,
|
|
193
|
+
cacheWriteCostPerMillion: 0,
|
|
194
|
+
cacheReadCostPerMillion: 0.26,
|
|
195
|
+
},
|
|
196
|
+
'zai/glm-5.3': {
|
|
197
|
+
inputCostPerMillion: 1.4,
|
|
198
|
+
outputCostPerMillion: 4.4,
|
|
199
|
+
cacheWriteCostPerMillion: 0,
|
|
200
|
+
cacheReadCostPerMillion: 0.26,
|
|
201
|
+
},
|
|
202
|
+
'glm-5.3-flash': {
|
|
203
|
+
inputCostPerMillion: 0.15,
|
|
204
|
+
outputCostPerMillion: 0.5,
|
|
205
|
+
cacheWriteCostPerMillion: 0,
|
|
206
|
+
cacheReadCostPerMillion: 0.03,
|
|
207
|
+
},
|
|
208
|
+
'zai/glm-5.3-flash': {
|
|
209
|
+
inputCostPerMillion: 0.15,
|
|
210
|
+
outputCostPerMillion: 0.5,
|
|
211
|
+
cacheWriteCostPerMillion: 0,
|
|
212
|
+
cacheReadCostPerMillion: 0.03,
|
|
213
|
+
},
|
|
214
|
+
'glm-5.3-flashx': {
|
|
215
|
+
inputCostPerMillion: 0.37,
|
|
216
|
+
outputCostPerMillion: 1.25,
|
|
217
|
+
cacheWriteCostPerMillion: 0,
|
|
218
|
+
cacheReadCostPerMillion: 0.075,
|
|
219
|
+
},
|
|
220
|
+
'zai/glm-5.3-flashx': {
|
|
221
|
+
inputCostPerMillion: 0.37,
|
|
222
|
+
outputCostPerMillion: 1.25,
|
|
223
|
+
cacheWriteCostPerMillion: 0,
|
|
224
|
+
cacheReadCostPerMillion: 0.075,
|
|
225
|
+
},
|
|
40
226
|
// ── Anthropic: Claude ──
|
|
41
227
|
// Fable 5.1 cache reads cost 0.025x input; Fable 5 retains its 0.1x rate.
|
|
42
228
|
'claude-fable-5-1': {
|
package/dist/modelContext.js
CHANGED
|
@@ -44,6 +44,38 @@ exports.getModelContextWindowSize = getModelContextWindowSize;
|
|
|
44
44
|
const modelAliases_1 = require("./modelAliases");
|
|
45
45
|
/** Known model context window sizes (in tokens); current models verified 2026-09-05. */
|
|
46
46
|
const MODEL_CONTEXT_SIZES = {
|
|
47
|
+
// Published API limits verified 2026-10-30; GLM's API documents 1M even
|
|
48
|
+
// though local weights may support more. Catalog/observed overrides win.
|
|
49
|
+
'gpt-6.1-sol': 1_050_000,
|
|
50
|
+
'openai/gpt-6.1-sol': 1_050_000,
|
|
51
|
+
'gpt-6.1-terra': 1_050_000,
|
|
52
|
+
'openai/gpt-6.1-terra': 1_050_000,
|
|
53
|
+
'claude-sonnet-5-5': 1_000_000,
|
|
54
|
+
'anthropic/claude-sonnet-5-5': 1_000_000,
|
|
55
|
+
'claude-sonnet-5.5': 1_000_000,
|
|
56
|
+
'anthropic/claude-sonnet-5.5': 1_000_000,
|
|
57
|
+
'claude-haiku-5-5': 1_000_000,
|
|
58
|
+
'anthropic/claude-haiku-5-5': 1_000_000,
|
|
59
|
+
'claude-haiku-5.5': 1_000_000,
|
|
60
|
+
'anthropic/claude-haiku-5.5': 1_000_000,
|
|
61
|
+
'gemini-4-argon': 1_048_576,
|
|
62
|
+
'gemini/gemini-4-argon': 1_048_576,
|
|
63
|
+
'gemini-3.8-flash': 1_048_576,
|
|
64
|
+
'gemini/gemini-3.8-flash': 1_048_576,
|
|
65
|
+
'deepseek-flash': 1_000_000,
|
|
66
|
+
'deepseek/deepseek-flash': 1_000_000,
|
|
67
|
+
'deepseek-v4-flash': 1_000_000,
|
|
68
|
+
'deepseek/deepseek-v4-flash': 1_000_000,
|
|
69
|
+
'deepseek-v4-flash-vision-exp': 1_000_000,
|
|
70
|
+
'deepseek/deepseek-v4-flash-vision-exp': 1_000_000,
|
|
71
|
+
'deepseek-v4-pro': 1_000_000,
|
|
72
|
+
'deepseek/deepseek-v4-pro': 1_000_000,
|
|
73
|
+
'glm-5.3': 1_000_000,
|
|
74
|
+
'zai/glm-5.3': 1_000_000,
|
|
75
|
+
'glm-5.3-flash': 1_000_000,
|
|
76
|
+
'zai/glm-5.3-flash': 1_000_000,
|
|
77
|
+
'glm-5.3-flashx': 1_000_000,
|
|
78
|
+
'zai/glm-5.3-flashx': 1_000_000,
|
|
47
79
|
// Claude — native 1M context (Fable 5+, Opus 4.6+, Sonnet 4.6+)
|
|
48
80
|
'claude-fable-5-1': 1_000_000,
|
|
49
81
|
'claude-fable-5.1': 1_000_000,
|
package/dist/modelInfo.js
CHANGED
|
@@ -60,6 +60,192 @@ Object.defineProperty(exports, "MODEL_PROVIDER_IDS", { enumerable: true, get: fu
|
|
|
60
60
|
* prefix matching cannot bridge the two, so both are needed.
|
|
61
61
|
*/
|
|
62
62
|
const PRICING_TABLE = {
|
|
63
|
+
// Current coding models verified 2026-10-30. Standard short-context rates;
|
|
64
|
+
// DeepSeek uses peak rates. Gemini promotions end 2026-11-30 (Argon) and
|
|
65
|
+
// 2026-12-31 (Flash). Haiku 5.5 costs more above 100K prompt tokens.
|
|
66
|
+
// Sources and estimation limits: docs/configuration/model-resolution.md.
|
|
67
|
+
// Explicit direct-provider keys support offline OpenCode model IDs without
|
|
68
|
+
// stripping arbitrary deployment prefixes or changing alias resolution.
|
|
69
|
+
'gpt-6.1-sol': {
|
|
70
|
+
inputCostPerMillion: 2,
|
|
71
|
+
outputCostPerMillion: 10,
|
|
72
|
+
cacheWriteCostPerMillion: 2.5,
|
|
73
|
+
cacheReadCostPerMillion: 0.1,
|
|
74
|
+
},
|
|
75
|
+
'openai/gpt-6.1-sol': {
|
|
76
|
+
inputCostPerMillion: 2,
|
|
77
|
+
outputCostPerMillion: 10,
|
|
78
|
+
cacheWriteCostPerMillion: 2.5,
|
|
79
|
+
cacheReadCostPerMillion: 0.1,
|
|
80
|
+
},
|
|
81
|
+
'gpt-6.1-terra': {
|
|
82
|
+
inputCostPerMillion: 1,
|
|
83
|
+
outputCostPerMillion: 5,
|
|
84
|
+
cacheWriteCostPerMillion: 1.25,
|
|
85
|
+
cacheReadCostPerMillion: 0.05,
|
|
86
|
+
},
|
|
87
|
+
'openai/gpt-6.1-terra': {
|
|
88
|
+
inputCostPerMillion: 1,
|
|
89
|
+
outputCostPerMillion: 5,
|
|
90
|
+
cacheWriteCostPerMillion: 1.25,
|
|
91
|
+
cacheReadCostPerMillion: 0.05,
|
|
92
|
+
},
|
|
93
|
+
'claude-sonnet-5-5': {
|
|
94
|
+
inputCostPerMillion: 2,
|
|
95
|
+
outputCostPerMillion: 10,
|
|
96
|
+
cacheWriteCostPerMillion: 2.5,
|
|
97
|
+
cacheReadCostPerMillion: 0.1,
|
|
98
|
+
},
|
|
99
|
+
'anthropic/claude-sonnet-5-5': {
|
|
100
|
+
inputCostPerMillion: 2,
|
|
101
|
+
outputCostPerMillion: 10,
|
|
102
|
+
cacheWriteCostPerMillion: 2.5,
|
|
103
|
+
cacheReadCostPerMillion: 0.1,
|
|
104
|
+
},
|
|
105
|
+
'claude-sonnet-5.5': {
|
|
106
|
+
inputCostPerMillion: 2,
|
|
107
|
+
outputCostPerMillion: 10,
|
|
108
|
+
cacheWriteCostPerMillion: 2.5,
|
|
109
|
+
cacheReadCostPerMillion: 0.1,
|
|
110
|
+
},
|
|
111
|
+
'anthropic/claude-sonnet-5.5': {
|
|
112
|
+
inputCostPerMillion: 2,
|
|
113
|
+
outputCostPerMillion: 10,
|
|
114
|
+
cacheWriteCostPerMillion: 2.5,
|
|
115
|
+
cacheReadCostPerMillion: 0.1,
|
|
116
|
+
},
|
|
117
|
+
'claude-haiku-5-5': {
|
|
118
|
+
inputCostPerMillion: 0.1,
|
|
119
|
+
outputCostPerMillion: 0.5,
|
|
120
|
+
cacheWriteCostPerMillion: 0.125,
|
|
121
|
+
cacheReadCostPerMillion: 0.01,
|
|
122
|
+
},
|
|
123
|
+
'anthropic/claude-haiku-5-5': {
|
|
124
|
+
inputCostPerMillion: 0.1,
|
|
125
|
+
outputCostPerMillion: 0.5,
|
|
126
|
+
cacheWriteCostPerMillion: 0.125,
|
|
127
|
+
cacheReadCostPerMillion: 0.01,
|
|
128
|
+
},
|
|
129
|
+
'claude-haiku-5.5': {
|
|
130
|
+
inputCostPerMillion: 0.1,
|
|
131
|
+
outputCostPerMillion: 0.5,
|
|
132
|
+
cacheWriteCostPerMillion: 0.125,
|
|
133
|
+
cacheReadCostPerMillion: 0.01,
|
|
134
|
+
},
|
|
135
|
+
'anthropic/claude-haiku-5.5': {
|
|
136
|
+
inputCostPerMillion: 0.1,
|
|
137
|
+
outputCostPerMillion: 0.5,
|
|
138
|
+
cacheWriteCostPerMillion: 0.125,
|
|
139
|
+
cacheReadCostPerMillion: 0.01,
|
|
140
|
+
},
|
|
141
|
+
'gemini-4-argon': {
|
|
142
|
+
inputCostPerMillion: 2,
|
|
143
|
+
outputCostPerMillion: 10,
|
|
144
|
+
cacheWriteCostPerMillion: 0,
|
|
145
|
+
cacheReadCostPerMillion: 0.1,
|
|
146
|
+
},
|
|
147
|
+
'gemini/gemini-4-argon': {
|
|
148
|
+
inputCostPerMillion: 2,
|
|
149
|
+
outputCostPerMillion: 10,
|
|
150
|
+
cacheWriteCostPerMillion: 0,
|
|
151
|
+
cacheReadCostPerMillion: 0.1,
|
|
152
|
+
},
|
|
153
|
+
'gemini-3.8-flash': {
|
|
154
|
+
inputCostPerMillion: 0.75,
|
|
155
|
+
outputCostPerMillion: 3.75,
|
|
156
|
+
cacheWriteCostPerMillion: 0,
|
|
157
|
+
cacheReadCostPerMillion: 0.075,
|
|
158
|
+
},
|
|
159
|
+
'gemini/gemini-3.8-flash': {
|
|
160
|
+
inputCostPerMillion: 0.75,
|
|
161
|
+
outputCostPerMillion: 3.75,
|
|
162
|
+
cacheWriteCostPerMillion: 0,
|
|
163
|
+
cacheReadCostPerMillion: 0.075,
|
|
164
|
+
},
|
|
165
|
+
'deepseek-flash': {
|
|
166
|
+
inputCostPerMillion: 0.3,
|
|
167
|
+
outputCostPerMillion: 1.2,
|
|
168
|
+
cacheWriteCostPerMillion: 0,
|
|
169
|
+
cacheReadCostPerMillion: 0.006,
|
|
170
|
+
},
|
|
171
|
+
'deepseek/deepseek-flash': {
|
|
172
|
+
inputCostPerMillion: 0.3,
|
|
173
|
+
outputCostPerMillion: 1.2,
|
|
174
|
+
cacheWriteCostPerMillion: 0,
|
|
175
|
+
cacheReadCostPerMillion: 0.006,
|
|
176
|
+
},
|
|
177
|
+
'deepseek-v4-flash': {
|
|
178
|
+
inputCostPerMillion: 0.3,
|
|
179
|
+
outputCostPerMillion: 1.2,
|
|
180
|
+
cacheWriteCostPerMillion: 0,
|
|
181
|
+
cacheReadCostPerMillion: 0.006,
|
|
182
|
+
},
|
|
183
|
+
'deepseek/deepseek-v4-flash': {
|
|
184
|
+
inputCostPerMillion: 0.3,
|
|
185
|
+
outputCostPerMillion: 1.2,
|
|
186
|
+
cacheWriteCostPerMillion: 0,
|
|
187
|
+
cacheReadCostPerMillion: 0.006,
|
|
188
|
+
},
|
|
189
|
+
'deepseek-v4-flash-vision-exp': {
|
|
190
|
+
inputCostPerMillion: 0.3,
|
|
191
|
+
outputCostPerMillion: 1.2,
|
|
192
|
+
cacheWriteCostPerMillion: 0,
|
|
193
|
+
cacheReadCostPerMillion: 0.006,
|
|
194
|
+
},
|
|
195
|
+
'deepseek/deepseek-v4-flash-vision-exp': {
|
|
196
|
+
inputCostPerMillion: 0.3,
|
|
197
|
+
outputCostPerMillion: 1.2,
|
|
198
|
+
cacheWriteCostPerMillion: 0,
|
|
199
|
+
cacheReadCostPerMillion: 0.006,
|
|
200
|
+
},
|
|
201
|
+
'deepseek-v4-pro': {
|
|
202
|
+
inputCostPerMillion: 1.32,
|
|
203
|
+
outputCostPerMillion: 3.96,
|
|
204
|
+
cacheWriteCostPerMillion: 0,
|
|
205
|
+
cacheReadCostPerMillion: 0.044,
|
|
206
|
+
},
|
|
207
|
+
'deepseek/deepseek-v4-pro': {
|
|
208
|
+
inputCostPerMillion: 1.32,
|
|
209
|
+
outputCostPerMillion: 3.96,
|
|
210
|
+
cacheWriteCostPerMillion: 0,
|
|
211
|
+
cacheReadCostPerMillion: 0.044,
|
|
212
|
+
},
|
|
213
|
+
'glm-5.3': {
|
|
214
|
+
inputCostPerMillion: 1.4,
|
|
215
|
+
outputCostPerMillion: 4.4,
|
|
216
|
+
cacheWriteCostPerMillion: 0,
|
|
217
|
+
cacheReadCostPerMillion: 0.26,
|
|
218
|
+
},
|
|
219
|
+
'zai/glm-5.3': {
|
|
220
|
+
inputCostPerMillion: 1.4,
|
|
221
|
+
outputCostPerMillion: 4.4,
|
|
222
|
+
cacheWriteCostPerMillion: 0,
|
|
223
|
+
cacheReadCostPerMillion: 0.26,
|
|
224
|
+
},
|
|
225
|
+
'glm-5.3-flash': {
|
|
226
|
+
inputCostPerMillion: 0.15,
|
|
227
|
+
outputCostPerMillion: 0.5,
|
|
228
|
+
cacheWriteCostPerMillion: 0,
|
|
229
|
+
cacheReadCostPerMillion: 0.03,
|
|
230
|
+
},
|
|
231
|
+
'zai/glm-5.3-flash': {
|
|
232
|
+
inputCostPerMillion: 0.15,
|
|
233
|
+
outputCostPerMillion: 0.5,
|
|
234
|
+
cacheWriteCostPerMillion: 0,
|
|
235
|
+
cacheReadCostPerMillion: 0.03,
|
|
236
|
+
},
|
|
237
|
+
'glm-5.3-flashx': {
|
|
238
|
+
inputCostPerMillion: 0.37,
|
|
239
|
+
outputCostPerMillion: 1.25,
|
|
240
|
+
cacheWriteCostPerMillion: 0,
|
|
241
|
+
cacheReadCostPerMillion: 0.075,
|
|
242
|
+
},
|
|
243
|
+
'zai/glm-5.3-flashx': {
|
|
244
|
+
inputCostPerMillion: 0.37,
|
|
245
|
+
outputCostPerMillion: 1.25,
|
|
246
|
+
cacheWriteCostPerMillion: 0,
|
|
247
|
+
cacheReadCostPerMillion: 0.075,
|
|
248
|
+
},
|
|
63
249
|
// ── Anthropic: Claude ──
|
|
64
250
|
// Fable 5.1 cache reads cost 0.025x input; Fable 5 retains its 0.1x rate.
|
|
65
251
|
'claude-fable-5-1': {
|