/ mlflow / utils / model_catalog / databricks.json
databricks.json
  1  {
  2    "schema_version": "1.0",
  3    "models": {
  4      "databricks-bge-large-en": {
  5        "mode": "embedding",
  6        "context_window": {
  7          "max_input": 512,
  8          "max_tokens": 512
  9        },
 10        "pricing": {
 11          "input_per_million_tokens": 0.10003,
 12          "output_per_million_tokens": 0.0
 13        },
 14        "capabilities": {
 15          "function_calling": false,
 16          "vision": false,
 17          "reasoning": false,
 18          "prompt_caching": false,
 19          "response_schema": false
 20        },
 21        "last_updated_at": "2026-04-24"
 22      },
 23      "databricks-claude-3-7-sonnet": {
 24        "mode": "chat",
 25        "context_window": {
 26          "max_input": 200000,
 27          "max_output": 128000,
 28          "max_tokens": 128000
 29        },
 30        "pricing": {
 31          "input_per_million_tokens": 2.99999,
 32          "output_per_million_tokens": 15.00002
 33        },
 34        "capabilities": {
 35          "function_calling": true,
 36          "vision": false,
 37          "reasoning": true,
 38          "prompt_caching": false,
 39          "response_schema": false
 40        },
 41        "last_updated_at": "2026-04-24"
 42      },
 43      "databricks-claude-haiku-4-5": {
 44        "mode": "chat",
 45        "context_window": {
 46          "max_input": 200000,
 47          "max_output": 64000,
 48          "max_tokens": 64000
 49        },
 50        "pricing": {
 51          "input_per_million_tokens": 1.00002,
 52          "output_per_million_tokens": 5.00003
 53        },
 54        "capabilities": {
 55          "function_calling": true,
 56          "vision": false,
 57          "reasoning": true,
 58          "prompt_caching": false,
 59          "response_schema": false
 60        },
 61        "last_updated_at": "2026-04-24"
 62      },
 63      "databricks-claude-opus-4": {
 64        "mode": "chat",
 65        "context_window": {
 66          "max_input": 200000,
 67          "max_output": 32000,
 68          "max_tokens": 32000
 69        },
 70        "pricing": {
 71          "input_per_million_tokens": 15.00002,
 72          "output_per_million_tokens": 75.00003
 73        },
 74        "capabilities": {
 75          "function_calling": true,
 76          "vision": false,
 77          "reasoning": true,
 78          "prompt_caching": false,
 79          "response_schema": false
 80        },
 81        "last_updated_at": "2026-04-24"
 82      },
 83      "databricks-claude-opus-4-1": {
 84        "mode": "chat",
 85        "context_window": {
 86          "max_input": 200000,
 87          "max_output": 32000,
 88          "max_tokens": 32000
 89        },
 90        "pricing": {
 91          "input_per_million_tokens": 15.00002,
 92          "output_per_million_tokens": 75.00003
 93        },
 94        "capabilities": {
 95          "function_calling": true,
 96          "vision": false,
 97          "reasoning": true,
 98          "prompt_caching": false,
 99          "response_schema": false
100        },
101        "last_updated_at": "2026-04-24"
102      },
103      "databricks-claude-opus-4-5": {
104        "mode": "chat",
105        "context_window": {
106          "max_input": 200000,
107          "max_output": 64000,
108          "max_tokens": 64000
109        },
110        "pricing": {
111          "input_per_million_tokens": 5.00003,
112          "output_per_million_tokens": 25.00001
113        },
114        "capabilities": {
115          "function_calling": true,
116          "vision": false,
117          "reasoning": true,
118          "prompt_caching": false,
119          "response_schema": false
120        },
121        "last_updated_at": "2026-04-24"
122      },
123      "databricks-claude-sonnet-4": {
124        "mode": "chat",
125        "context_window": {
126          "max_input": 200000,
127          "max_output": 64000,
128          "max_tokens": 64000
129        },
130        "pricing": {
131          "input_per_million_tokens": 2.99999,
132          "output_per_million_tokens": 15.00002
133        },
134        "capabilities": {
135          "function_calling": true,
136          "vision": false,
137          "reasoning": true,
138          "prompt_caching": false,
139          "response_schema": false
140        },
141        "last_updated_at": "2026-04-24"
142      },
143      "databricks-claude-sonnet-4-1": {
144        "mode": "chat",
145        "context_window": {
146          "max_input": 200000,
147          "max_output": 64000,
148          "max_tokens": 64000
149        },
150        "pricing": {
151          "input_per_million_tokens": 2.99999,
152          "output_per_million_tokens": 15.00002
153        },
154        "capabilities": {
155          "function_calling": true,
156          "vision": false,
157          "reasoning": true,
158          "prompt_caching": false,
159          "response_schema": false
160        },
161        "last_updated_at": "2026-04-24"
162      },
163      "databricks-claude-sonnet-4-5": {
164        "mode": "chat",
165        "context_window": {
166          "max_input": 200000,
167          "max_output": 64000,
168          "max_tokens": 64000
169        },
170        "pricing": {
171          "input_per_million_tokens": 2.99999,
172          "output_per_million_tokens": 15.00002
173        },
174        "capabilities": {
175          "function_calling": true,
176          "vision": false,
177          "reasoning": true,
178          "prompt_caching": false,
179          "response_schema": false
180        },
181        "last_updated_at": "2026-04-24"
182      },
183      "databricks-gemini-2-5-flash": {
184        "mode": "chat",
185        "context_window": {
186          "max_input": 1048576,
187          "max_output": 65535,
188          "max_tokens": 65535
189        },
190        "pricing": {
191          "input_per_million_tokens": 0.30002,
192          "output_per_million_tokens": 2.49998
193        },
194        "capabilities": {
195          "function_calling": true,
196          "vision": false,
197          "reasoning": false,
198          "prompt_caching": false,
199          "response_schema": false
200        },
201        "last_updated_at": "2026-04-24"
202      },
203      "databricks-gemini-2-5-pro": {
204        "mode": "chat",
205        "context_window": {
206          "max_input": 1048576,
207          "max_output": 65536,
208          "max_tokens": 65536
209        },
210        "pricing": {
211          "input_per_million_tokens": 1.24999,
212          "output_per_million_tokens": 9.99999
213        },
214        "capabilities": {
215          "function_calling": true,
216          "vision": false,
217          "reasoning": false,
218          "prompt_caching": false,
219          "response_schema": false
220        },
221        "last_updated_at": "2026-04-24"
222      },
223      "databricks-gemma-3-12b": {
224        "mode": "chat",
225        "context_window": {
226          "max_input": 128000,
227          "max_output": 32000,
228          "max_tokens": 32000
229        },
230        "pricing": {
231          "input_per_million_tokens": 0.15001,
232          "output_per_million_tokens": 0.50001
233        },
234        "capabilities": {
235          "function_calling": false,
236          "vision": false,
237          "reasoning": false,
238          "prompt_caching": false,
239          "response_schema": false
240        },
241        "last_updated_at": "2026-04-24"
242      },
243      "databricks-gpt-5": {
244        "mode": "chat",
245        "context_window": {
246          "max_input": 272000,
247          "max_output": 128000,
248          "max_tokens": 128000
249        },
250        "pricing": {
251          "input_per_million_tokens": 1.24999,
252          "output_per_million_tokens": 9.99999
253        },
254        "capabilities": {
255          "function_calling": false,
256          "vision": false,
257          "reasoning": false,
258          "prompt_caching": false,
259          "response_schema": false
260        },
261        "last_updated_at": "2026-04-24"
262      },
263      "databricks-gpt-5-1": {
264        "mode": "chat",
265        "context_window": {
266          "max_input": 272000,
267          "max_output": 128000,
268          "max_tokens": 128000
269        },
270        "pricing": {
271          "input_per_million_tokens": 1.24999,
272          "output_per_million_tokens": 9.99999
273        },
274        "capabilities": {
275          "function_calling": false,
276          "vision": false,
277          "reasoning": false,
278          "prompt_caching": false,
279          "response_schema": false
280        },
281        "last_updated_at": "2026-04-24"
282      },
283      "databricks-gpt-5-mini": {
284        "mode": "chat",
285        "context_window": {
286          "max_input": 272000,
287          "max_output": 128000,
288          "max_tokens": 128000
289        },
290        "pricing": {
291          "input_per_million_tokens": 0.24997,
292          "output_per_million_tokens": 1.99997
293        },
294        "capabilities": {
295          "function_calling": false,
296          "vision": false,
297          "reasoning": false,
298          "prompt_caching": false,
299          "response_schema": false
300        },
301        "last_updated_at": "2026-04-24"
302      },
303      "databricks-gpt-5-nano": {
304        "mode": "chat",
305        "context_window": {
306          "max_input": 272000,
307          "max_output": 128000,
308          "max_tokens": 128000
309        },
310        "pricing": {
311          "input_per_million_tokens": 0.04998,
312          "output_per_million_tokens": 0.39998
313        },
314        "capabilities": {
315          "function_calling": false,
316          "vision": false,
317          "reasoning": false,
318          "prompt_caching": false,
319          "response_schema": false
320        },
321        "last_updated_at": "2026-04-24"
322      },
323      "databricks-gpt-oss-120b": {
324        "mode": "chat",
325        "context_window": {
326          "max_input": 131072,
327          "max_output": 131072,
328          "max_tokens": 131072
329        },
330        "pricing": {
331          "input_per_million_tokens": 0.15001,
332          "output_per_million_tokens": 0.59997
333        },
334        "capabilities": {
335          "function_calling": false,
336          "vision": false,
337          "reasoning": false,
338          "prompt_caching": false,
339          "response_schema": false
340        },
341        "last_updated_at": "2026-04-24"
342      },
343      "databricks-gpt-oss-20b": {
344        "mode": "chat",
345        "context_window": {
346          "max_input": 131072,
347          "max_output": 131072,
348          "max_tokens": 131072
349        },
350        "pricing": {
351          "input_per_million_tokens": 0.07,
352          "output_per_million_tokens": 0.30002
353        },
354        "capabilities": {
355          "function_calling": false,
356          "vision": false,
357          "reasoning": false,
358          "prompt_caching": false,
359          "response_schema": false
360        },
361        "last_updated_at": "2026-04-24"
362      },
363      "databricks-gte-large-en": {
364        "mode": "embedding",
365        "context_window": {
366          "max_input": 8192,
367          "max_tokens": 8192
368        },
369        "pricing": {
370          "input_per_million_tokens": 0.12999,
371          "output_per_million_tokens": 0.0
372        },
373        "capabilities": {
374          "function_calling": false,
375          "vision": false,
376          "reasoning": false,
377          "prompt_caching": false,
378          "response_schema": false
379        },
380        "last_updated_at": "2026-04-24"
381      },
382      "databricks-llama-2-70b-chat": {
383        "mode": "chat",
384        "context_window": {
385          "max_input": 4096,
386          "max_output": 4096,
387          "max_tokens": 4096
388        },
389        "pricing": {
390          "input_per_million_tokens": 0.50001,
391          "output_per_million_tokens": 1.50003
392        },
393        "capabilities": {
394          "function_calling": false,
395          "vision": false,
396          "reasoning": false,
397          "prompt_caching": false,
398          "response_schema": false
399        },
400        "last_updated_at": "2026-04-24"
401      },
402      "databricks-llama-4-maverick": {
403        "mode": "chat",
404        "context_window": {
405          "max_input": 128000,
406          "max_output": 128000,
407          "max_tokens": 128000
408        },
409        "pricing": {
410          "input_per_million_tokens": 0.50001,
411          "output_per_million_tokens": 1.50003
412        },
413        "capabilities": {
414          "function_calling": false,
415          "vision": false,
416          "reasoning": false,
417          "prompt_caching": false,
418          "response_schema": false
419        },
420        "last_updated_at": "2026-04-24"
421      },
422      "databricks-meta-llama-3-1-405b-instruct": {
423        "mode": "chat",
424        "context_window": {
425          "max_input": 128000,
426          "max_output": 128000,
427          "max_tokens": 128000
428        },
429        "pricing": {
430          "input_per_million_tokens": 5.00003,
431          "output_per_million_tokens": 15.00002
432        },
433        "capabilities": {
434          "function_calling": false,
435          "vision": false,
436          "reasoning": false,
437          "prompt_caching": false,
438          "response_schema": false
439        },
440        "last_updated_at": "2026-04-24"
441      },
442      "databricks-meta-llama-3-1-8b-instruct": {
443        "mode": "chat",
444        "context_window": {
445          "max_input": 200000,
446          "max_output": 128000,
447          "max_tokens": 128000
448        },
449        "pricing": {
450          "input_per_million_tokens": 0.15001,
451          "output_per_million_tokens": 0.45003
452        },
453        "capabilities": {
454          "function_calling": false,
455          "vision": false,
456          "reasoning": false,
457          "prompt_caching": false,
458          "response_schema": false
459        },
460        "last_updated_at": "2026-04-24"
461      },
462      "databricks-meta-llama-3-3-70b-instruct": {
463        "mode": "chat",
464        "context_window": {
465          "max_input": 128000,
466          "max_output": 128000,
467          "max_tokens": 128000
468        },
469        "pricing": {
470          "input_per_million_tokens": 0.50001,
471          "output_per_million_tokens": 1.50003
472        },
473        "capabilities": {
474          "function_calling": false,
475          "vision": false,
476          "reasoning": false,
477          "prompt_caching": false,
478          "response_schema": false
479        },
480        "last_updated_at": "2026-04-24"
481      },
482      "databricks-meta-llama-3-70b-instruct": {
483        "mode": "chat",
484        "context_window": {
485          "max_input": 128000,
486          "max_output": 128000,
487          "max_tokens": 128000
488        },
489        "pricing": {
490          "input_per_million_tokens": 1.00002,
491          "output_per_million_tokens": 2.99999
492        },
493        "capabilities": {
494          "function_calling": false,
495          "vision": false,
496          "reasoning": false,
497          "prompt_caching": false,
498          "response_schema": false
499        },
500        "last_updated_at": "2026-04-24"
501      },
502      "databricks-mixtral-8x7b-instruct": {
503        "mode": "chat",
504        "context_window": {
505          "max_input": 4096,
506          "max_output": 4096,
507          "max_tokens": 4096
508        },
509        "pricing": {
510          "input_per_million_tokens": 0.50001,
511          "output_per_million_tokens": 1.00002
512        },
513        "capabilities": {
514          "function_calling": false,
515          "vision": false,
516          "reasoning": false,
517          "prompt_caching": false,
518          "response_schema": false
519        },
520        "last_updated_at": "2026-04-24"
521      },
522      "databricks-mpt-30b-instruct": {
523        "mode": "chat",
524        "context_window": {
525          "max_input": 8192,
526          "max_output": 8192,
527          "max_tokens": 8192
528        },
529        "pricing": {
530          "input_per_million_tokens": 1.00002,
531          "output_per_million_tokens": 1.00002
532        },
533        "capabilities": {
534          "function_calling": false,
535          "vision": false,
536          "reasoning": false,
537          "prompt_caching": false,
538          "response_schema": false
539        },
540        "last_updated_at": "2026-04-24"
541      },
542      "databricks-mpt-7b-instruct": {
543        "mode": "chat",
544        "context_window": {
545          "max_input": 8192,
546          "max_output": 8192,
547          "max_tokens": 8192
548        },
549        "pricing": {
550          "input_per_million_tokens": 0.50001,
551          "output_per_million_tokens": 0.0
552        },
553        "capabilities": {
554          "function_calling": false,
555          "vision": false,
556          "reasoning": false,
557          "prompt_caching": false,
558          "response_schema": false
559        },
560        "last_updated_at": "2026-04-24"
561      }
562    }
563  }