1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
use super::*;
/// Profile payloads for non-OpenAI models exposed through OpenAI-compatible
/// APIs (NVIDIA NIM, Alibaba, MiniMax, Mistral, Moonshot, xAI, Microsoft). Pure value
/// store keyed by canonical id; which provider surfaces each model is offered
/// under is decided by `REGISTRY`, not here. Sourced from models.dev unless
/// noted otherwise.
///
/// The id is lowercased so the cased canonical ids and aliases below resolve
/// regardless of how the caller passed them.
///
/// `structured_output` is set to `false` where the upstream models.dev entry
/// does not assert it: absence of the field is not a claim of support, so we
/// do not advertise a capability we cannot confirm.
pub(super) fn third_party_profile_data(model_id: &str) -> Option<ModelProfile> {
match model_id.to_ascii_lowercase().as_str() {
// NVIDIA Nemotron 3 Super — flagship Nemotron reasoning model.
// Source: models.dev (nvidia provider).
"nemotron-3-super-120b-a12b" | "nvidia/nemotron-3-super-120b-a12b" => Some(ModelProfile {
name: "Nemotron 3 Super".into(),
family: "nemotron-3-super".into(),
description: None,
release_date: Some("2026-03-11".into()),
last_updated: Some("2026-03-11".into()),
attachment: false,
reasoning: true,
temperature: true,
knowledge: Some("2024-04-01".into()),
tool_call: true,
structured_output: false,
open_weights: true,
cost: Some(ModelCost {
input: 0.20,
output: 0.80,
cache_read: None,
cache_write: None,
cost_tiers: vec![],
}),
limits: Some(ModelLimits {
context: 262_144,
input: None,
output: 262_144,
max_media: None,
}),
modalities: Some(ModelModalities {
input: vec![Modality::Text],
output: vec![Modality::Text],
}),
reasoning_effort: None,
speed: None,
verbosity: None,
tool_search: false,
supported_parameters: Vec::new(),
supports_phases: false,
supports_server_compaction: false,
decisions: None,
}),
// Alibaba Qwen3.7 Max — flagship Qwen model.
// Source: models.dev (alibaba provider). Knowledge cutoff not published.
"qwen3.7-max" | "qwen/qwen3.7-max" => Some(ModelProfile {
name: "Qwen3.7 Max".into(),
family: "qwen3.7-max".into(),
description: None,
release_date: Some("2026-05-21".into()),
last_updated: Some("2026-05-21".into()),
attachment: false,
reasoning: true,
temperature: true,
knowledge: None,
tool_call: true,
structured_output: false,
open_weights: false,
cost: Some(ModelCost {
input: 2.50,
output: 7.50,
cache_read: Some(0.50),
cache_write: None,
cost_tiers: vec![],
}),
limits: Some(ModelLimits {
context: 1_000_000,
input: None,
output: 65_536,
max_media: None,
}),
modalities: Some(ModelModalities {
input: vec![Modality::Text],
output: vec![Modality::Text],
}),
reasoning_effort: None,
speed: None,
verbosity: None,
tool_search: false,
supported_parameters: Vec::new(),
supports_phases: false,
supports_server_compaction: false,
decisions: None,
}),
// Microsoft MAI-1-preview — Microsoft's first end-to-end in-house
// foundation model. Not on models.dev; sourced from Microsoft's official
// announcement (microsoft.ai/news/two-new-in-house-models). It is a
// text-only instruction model (not a reasoning model); context window,
// pricing, and knowledge cutoff were never publicly disclosed, so cost
// and limits are left unset rather than guessed.
"mai-1-preview" | "microsoft/mai-1-preview" => Some(ModelProfile {
name: "MAI-1-preview".into(),
family: "mai-1-preview".into(),
description: None,
release_date: Some("2025-08-28".into()),
last_updated: None,
attachment: false,
reasoning: false,
temperature: true,
knowledge: None,
tool_call: false,
structured_output: false,
open_weights: false,
cost: None,
limits: None,
modalities: Some(ModelModalities {
input: vec![Modality::Text],
output: vec![Modality::Text],
}),
reasoning_effort: None,
speed: None,
verbosity: None,
tool_search: false,
supported_parameters: Vec::new(),
supports_phases: false,
supports_server_compaction: false,
decisions: None,
}),
// Microsoft MAI-Code-1-Flash — Microsoft's in-house, latency-optimized
// coding model served via Azure AI Foundry (and OpenAI-compatible
// gateways). It is a fast, tool-calling code model rather than a graded
// reasoning model. Microsoft has not published pricing, context window,
// or knowledge cutoff, so cost/limits are left unset rather than guessed
// (same policy as MAI-1-preview).
"mai-code-1-flash" | "microsoft/mai-code-1-flash" => Some(ModelProfile {
name: "MAI-Code-1-Flash".into(),
family: "mai-code-1".into(),
description: Some(
"Microsoft's latency-optimized in-house coding model (Azure AI Foundry).".into(),
),
release_date: None,
last_updated: None,
attachment: false,
reasoning: false,
temperature: true,
knowledge: None,
tool_call: true,
structured_output: false,
open_weights: false,
cost: None,
limits: None,
modalities: Some(ModelModalities {
input: vec![Modality::Text],
output: vec![Modality::Text],
}),
reasoning_effort: None,
speed: None,
verbosity: None,
tool_search: false,
supported_parameters: Vec::new(),
supports_phases: false,
supports_server_compaction: false,
decisions: None,
}),
// MiniMax-M3 — flagship MiniMax model. Source: models.dev (minimax
// provider). Reasoning is a toggle upstream (no graded effort).
"minimax-m3" | "minimax/minimax-m3" => Some(ModelProfile {
name: "MiniMax-M3".into(),
family: "minimax-m3".into(),
description: None,
release_date: Some("2026-06-01".into()),
last_updated: Some("2026-06-01".into()),
attachment: true,
reasoning: true,
temperature: true,
knowledge: None,
tool_call: true,
structured_output: false,
open_weights: true,
cost: Some(ModelCost {
input: 0.60,
output: 2.40,
cache_read: Some(0.12),
cache_write: None,
cost_tiers: vec![],
}),
limits: Some(ModelLimits {
context: 512_000,
input: None,
output: 128_000,
max_media: None,
}),
modalities: Some(ModelModalities {
input: vec![Modality::Text, Modality::Image, Modality::Video],
output: vec![Modality::Text],
}),
reasoning_effort: None,
speed: None,
verbosity: None,
tool_search: false,
supported_parameters: Vec::new(),
supports_phases: false,
supports_server_compaction: false,
decisions: None,
}),
// Moonshot Kimi K2 Thinking — flagship Kimi reasoning model.
// Source: models.dev (moonshotai provider).
"kimi-k2-thinking" | "moonshotai/kimi-k2-thinking" => Some(ModelProfile {
name: "Kimi K2 Thinking".into(),
family: "kimi-k2-thinking".into(),
description: None,
release_date: Some("2025-11-06".into()),
last_updated: Some("2025-11-06".into()),
attachment: false,
reasoning: true,
temperature: true,
knowledge: Some("2024-08-01".into()),
tool_call: true,
structured_output: false,
open_weights: true,
cost: Some(ModelCost {
input: 0.60,
output: 2.50,
cache_read: Some(0.15),
cache_write: None,
cost_tiers: vec![],
}),
limits: Some(ModelLimits {
context: 262_144,
input: None,
output: 262_144,
max_media: None,
}),
modalities: Some(ModelModalities {
input: vec![Modality::Text],
output: vec![Modality::Text],
}),
reasoning_effort: None,
speed: None,
verbosity: None,
tool_search: false,
supported_parameters: Vec::new(),
supports_phases: false,
supports_server_compaction: false,
decisions: None,
}),
// Moonshot Kimi K3 — flagship multimodal Kimi model with a 1M-token
// context window. Source: models.dev (moonshotai provider). models.dev
// lists graded reasoning effort (low/high/max), but like the other
// OpenAI-compatible third-party models here we don't wire a graded
// effort selector; `reasoning: true` still gates reasoning support.
// Knowledge cutoff not published.
"kimi-k3" | "moonshotai/kimi-k3" => Some(ModelProfile {
name: "Kimi K3".into(),
family: "kimi-k3".into(),
description: None,
release_date: Some("2026-07-16".into()),
last_updated: Some("2026-07-16".into()),
attachment: true,
reasoning: true,
temperature: false,
knowledge: None,
tool_call: true,
structured_output: true,
open_weights: true,
cost: Some(ModelCost {
input: 3.00,
output: 15.00,
cache_read: Some(0.30),
cache_write: None,
cost_tiers: vec![],
}),
limits: Some(ModelLimits {
context: 1_048_576,
input: None,
output: 131_072,
max_media: None,
}),
modalities: Some(ModelModalities {
input: vec![Modality::Text, Modality::Image, Modality::Video],
output: vec![Modality::Text],
}),
reasoning_effort: None,
speed: None,
verbosity: None,
tool_search: false,
supported_parameters: Vec::new(),
supports_phases: false,
supports_server_compaction: false,
decisions: None,
}),
// Mistral Large 4: Mistral's 1T-parameter (49B active) multimodal MoE,
// a hybrid instruct-and-reasoning model, public preview 2026-10-06.
// Source: models.dev (mistral provider) for limits and capabilities,
// checked against Mistral's own `/v1/models` (524,288 context, vision,
// function calling, reasoning) and the launch post
// (mistral.ai/news/mistral-large-4) for list price: $1.36 / $4.18.
// models.dev's mistral entry carries $0.68 / $2.09, which is the
// launch discount (`/v1/models` bills it as
// `mistral-large-4-0-launch-discount`); the table keeps list price so
// estimates do not silently halve when the promotion ends. Cache read
// is models.dev's 10% of input at list price.
//
// Reasoning is a toggle, not a grade: the API accepts only
// `reasoning_effort` "none" and "high" (anything else is a 400), and
// omitting it answers without thinking. Weights are promised for the
// end of October 2026 under an unannounced license, so `open_weights`
// stays false until they ship. Knowledge cutoff not published.
"mistral-large-4" => Some(ModelProfile {
name: "Mistral Large 4".into(),
family: "mistral-large".into(),
description: None,
release_date: Some("2026-10-06".into()),
last_updated: Some("2026-10-06".into()),
attachment: true,
reasoning: true,
temperature: true,
knowledge: None,
tool_call: true,
structured_output: true,
open_weights: false,
cost: Some(ModelCost {
input: 1.36,
output: 4.18,
cache_read: Some(0.14),
cache_write: None,
cost_tiers: vec![],
}),
limits: Some(ModelLimits {
context: 524_288,
input: None,
output: 262_144,
max_media: None,
}),
modalities: Some(ModelModalities {
input: vec![Modality::Text, Modality::Image],
output: vec![Modality::Text],
}),
reasoning_effort: Some(reasoning_effort_toggle()),
speed: None,
verbosity: None,
tool_search: false,
supported_parameters: Vec::new(),
supports_phases: false,
supports_server_compaction: false,
decisions: None,
}),
// xAI Grok 4.3 — flagship Grok model. Source: models.dev (xai provider).
// Pricing has a >200K-token tier. Knowledge cutoff not published.
"grok-4.3" | "x-ai/grok-4.3" | "xai/grok-4.3" => Some(ModelProfile {
name: "Grok 4.3".into(),
family: "grok-4.3".into(),
description: None,
release_date: Some("2026-04-17".into()),
last_updated: Some("2026-04-17".into()),
attachment: true,
reasoning: true,
temperature: true,
knowledge: None,
tool_call: true,
structured_output: false,
open_weights: false,
cost: Some(ModelCost {
input: 1.25,
output: 2.50,
cache_read: Some(0.20),
cache_write: None,
cost_tiers: vec![CostTier {
above_tokens: 200_000,
input: 2.50,
output: 5.00,
cache_read: Some(0.40),
cache_write: None,
}],
}),
limits: Some(ModelLimits {
context: 1_000_000,
input: None,
output: 30_000,
max_media: None,
}),
modalities: Some(ModelModalities {
input: vec![Modality::Text, Modality::Image, Modality::Pdf],
output: vec![Modality::Text],
}),
reasoning_effort: None,
speed: None,
verbosity: None,
tool_search: false,
supported_parameters: Vec::new(),
supports_phases: false,
supports_server_compaction: false,
decisions: None,
}),
_ => None,
}
}