1{
2 "name": "Venice AI",
3 "id": "venice",
4 "api_key": "$VENICE_API_KEY",
5 "api_endpoint": "https://api.venice.ai/api/v1",
6 "type": "openai-compat",
7 "default_large_model_id": "claude-opus-4-6",
8 "default_small_model_id": "qwen3-5-35b-a3b",
9 "models": [
10 {
11 "id": "claude-opus-4-5",
12 "name": "Claude Opus 4.5",
13 "cost_per_1m_in": 6,
14 "cost_per_1m_out": 30,
15 "cost_per_1m_in_cached": 0,
16 "cost_per_1m_out_cached": 0,
17 "context_window": 198000,
18 "default_max_tokens": 32768,
19 "can_reason": true,
20 "reasoning_levels": [
21 "low",
22 "medium",
23 "high"
24 ],
25 "default_reasoning_effort": "medium",
26 "supports_attachments": true,
27 "options": {}
28 },
29 {
30 "id": "claude-opus-4-6",
31 "name": "Claude Opus 4.6",
32 "cost_per_1m_in": 6,
33 "cost_per_1m_out": 30,
34 "cost_per_1m_in_cached": 0,
35 "cost_per_1m_out_cached": 0,
36 "context_window": 1000000,
37 "default_max_tokens": 32768,
38 "can_reason": true,
39 "reasoning_levels": [
40 "low",
41 "medium",
42 "high"
43 ],
44 "default_reasoning_effort": "medium",
45 "supports_attachments": true,
46 "options": {}
47 },
48 {
49 "id": "claude-sonnet-4-5",
50 "name": "Claude Sonnet 4.5",
51 "cost_per_1m_in": 3.75,
52 "cost_per_1m_out": 18.75,
53 "cost_per_1m_in_cached": 0,
54 "cost_per_1m_out_cached": 0,
55 "context_window": 198000,
56 "default_max_tokens": 32768,
57 "can_reason": true,
58 "reasoning_levels": [
59 "low",
60 "medium",
61 "high"
62 ],
63 "default_reasoning_effort": "medium",
64 "supports_attachments": true,
65 "options": {}
66 },
67 {
68 "id": "claude-sonnet-4-6",
69 "name": "Claude Sonnet 4.6",
70 "cost_per_1m_in": 3.6,
71 "cost_per_1m_out": 18,
72 "cost_per_1m_in_cached": 0,
73 "cost_per_1m_out_cached": 0,
74 "context_window": 1000000,
75 "default_max_tokens": 32768,
76 "can_reason": true,
77 "reasoning_levels": [
78 "low",
79 "medium",
80 "high"
81 ],
82 "default_reasoning_effort": "medium",
83 "supports_attachments": true,
84 "options": {}
85 },
86 {
87 "id": "zai-org-glm-4.6",
88 "name": "GLM 4.6",
89 "cost_per_1m_in": 0.85,
90 "cost_per_1m_out": 2.75,
91 "cost_per_1m_in_cached": 0,
92 "cost_per_1m_out_cached": 0,
93 "context_window": 198000,
94 "default_max_tokens": 32768,
95 "can_reason": false,
96 "supports_attachments": false,
97 "options": {}
98 },
99 {
100 "id": "zai-org-glm-4.7",
101 "name": "GLM 4.7",
102 "cost_per_1m_in": 0.55,
103 "cost_per_1m_out": 2.65,
104 "cost_per_1m_in_cached": 0,
105 "cost_per_1m_out_cached": 0,
106 "context_window": 198000,
107 "default_max_tokens": 32768,
108 "can_reason": true,
109 "reasoning_levels": [
110 "low",
111 "medium",
112 "high"
113 ],
114 "default_reasoning_effort": "medium",
115 "supports_attachments": false,
116 "options": {}
117 },
118 {
119 "id": "zai-org-glm-4.7-flash",
120 "name": "GLM 4.7 Flash",
121 "cost_per_1m_in": 0.125,
122 "cost_per_1m_out": 0.5,
123 "cost_per_1m_in_cached": 0,
124 "cost_per_1m_out_cached": 0,
125 "context_window": 128000,
126 "default_max_tokens": 32000,
127 "can_reason": true,
128 "reasoning_levels": [
129 "low",
130 "medium",
131 "high"
132 ],
133 "default_reasoning_effort": "medium",
134 "supports_attachments": false,
135 "options": {}
136 },
137 {
138 "id": "olafangensan-glm-4.7-flash-heretic",
139 "name": "GLM 4.7 Flash Heretic",
140 "cost_per_1m_in": 0.14,
141 "cost_per_1m_out": 0.8,
142 "cost_per_1m_in_cached": 0,
143 "cost_per_1m_out_cached": 0,
144 "context_window": 128000,
145 "default_max_tokens": 32000,
146 "can_reason": true,
147 "reasoning_levels": [
148 "low",
149 "medium",
150 "high"
151 ],
152 "default_reasoning_effort": "medium",
153 "supports_attachments": false,
154 "options": {}
155 },
156 {
157 "id": "zai-org-glm-5",
158 "name": "GLM 5",
159 "cost_per_1m_in": 1,
160 "cost_per_1m_out": 3.2,
161 "cost_per_1m_in_cached": 0,
162 "cost_per_1m_out_cached": 0,
163 "context_window": 198000,
164 "default_max_tokens": 32768,
165 "can_reason": true,
166 "reasoning_levels": [
167 "low",
168 "medium",
169 "high"
170 ],
171 "default_reasoning_effort": "medium",
172 "supports_attachments": false,
173 "options": {}
174 },
175 {
176 "id": "openai-gpt-4o-2024-11-20",
177 "name": "GPT-4o",
178 "cost_per_1m_in": 3.125,
179 "cost_per_1m_out": 12.5,
180 "cost_per_1m_in_cached": 0,
181 "cost_per_1m_out_cached": 0,
182 "context_window": 128000,
183 "default_max_tokens": 32000,
184 "can_reason": false,
185 "supports_attachments": true,
186 "options": {}
187 },
188 {
189 "id": "openai-gpt-4o-mini-2024-07-18",
190 "name": "GPT-4o Mini",
191 "cost_per_1m_in": 0.1875,
192 "cost_per_1m_out": 0.75,
193 "cost_per_1m_in_cached": 0,
194 "cost_per_1m_out_cached": 0,
195 "context_window": 128000,
196 "default_max_tokens": 32000,
197 "can_reason": false,
198 "supports_attachments": true,
199 "options": {}
200 },
201 {
202 "id": "openai-gpt-52",
203 "name": "GPT-5.2",
204 "cost_per_1m_in": 2.19,
205 "cost_per_1m_out": 17.5,
206 "cost_per_1m_in_cached": 0,
207 "cost_per_1m_out_cached": 0,
208 "context_window": 256000,
209 "default_max_tokens": 32768,
210 "can_reason": true,
211 "reasoning_levels": [
212 "low",
213 "medium",
214 "high"
215 ],
216 "default_reasoning_effort": "medium",
217 "supports_attachments": false,
218 "options": {}
219 },
220 {
221 "id": "openai-gpt-52-codex",
222 "name": "GPT-5.2 Codex",
223 "cost_per_1m_in": 2.19,
224 "cost_per_1m_out": 17.5,
225 "cost_per_1m_in_cached": 0,
226 "cost_per_1m_out_cached": 0,
227 "context_window": 256000,
228 "default_max_tokens": 32768,
229 "can_reason": true,
230 "reasoning_levels": [
231 "low",
232 "medium",
233 "high"
234 ],
235 "default_reasoning_effort": "medium",
236 "supports_attachments": true,
237 "options": {}
238 },
239 {
240 "id": "openai-gpt-53-codex",
241 "name": "GPT-5.3 Codex",
242 "cost_per_1m_in": 2.19,
243 "cost_per_1m_out": 17.5,
244 "cost_per_1m_in_cached": 0,
245 "cost_per_1m_out_cached": 0,
246 "context_window": 400000,
247 "default_max_tokens": 32768,
248 "can_reason": true,
249 "reasoning_levels": [
250 "low",
251 "medium",
252 "high"
253 ],
254 "default_reasoning_effort": "medium",
255 "supports_attachments": true,
256 "options": {}
257 },
258 {
259 "id": "openai-gpt-54",
260 "name": "GPT-5.4",
261 "cost_per_1m_in": 3.13,
262 "cost_per_1m_out": 18.8,
263 "cost_per_1m_in_cached": 0,
264 "cost_per_1m_out_cached": 0,
265 "context_window": 1000000,
266 "default_max_tokens": 32768,
267 "can_reason": true,
268 "reasoning_levels": [
269 "low",
270 "medium",
271 "high"
272 ],
273 "default_reasoning_effort": "medium",
274 "supports_attachments": false,
275 "options": {}
276 },
277 {
278 "id": "gemini-3-flash-preview",
279 "name": "Gemini 3 Flash Preview",
280 "cost_per_1m_in": 0.7,
281 "cost_per_1m_out": 3.75,
282 "cost_per_1m_in_cached": 0,
283 "cost_per_1m_out_cached": 0,
284 "context_window": 256000,
285 "default_max_tokens": 32768,
286 "can_reason": true,
287 "reasoning_levels": [
288 "low",
289 "medium",
290 "high"
291 ],
292 "default_reasoning_effort": "medium",
293 "supports_attachments": true,
294 "options": {}
295 },
296 {
297 "id": "gemini-3-pro-preview",
298 "name": "Gemini 3 Pro Preview",
299 "cost_per_1m_in": 2.5,
300 "cost_per_1m_out": 15,
301 "cost_per_1m_in_cached": 0,
302 "cost_per_1m_out_cached": 0,
303 "context_window": 198000,
304 "default_max_tokens": 32768,
305 "can_reason": true,
306 "reasoning_levels": [
307 "low",
308 "medium",
309 "high"
310 ],
311 "default_reasoning_effort": "medium",
312 "supports_attachments": true,
313 "options": {}
314 },
315 {
316 "id": "gemini-3-1-pro-preview",
317 "name": "Gemini 3.1 Pro Preview",
318 "cost_per_1m_in": 2.5,
319 "cost_per_1m_out": 15,
320 "cost_per_1m_in_cached": 0,
321 "cost_per_1m_out_cached": 0,
322 "context_window": 1000000,
323 "default_max_tokens": 32768,
324 "can_reason": true,
325 "reasoning_levels": [
326 "low",
327 "medium",
328 "high"
329 ],
330 "default_reasoning_effort": "medium",
331 "supports_attachments": true,
332 "options": {}
333 },
334 {
335 "id": "google-gemma-3-27b-it",
336 "name": "Google Gemma 3 27B Instruct",
337 "cost_per_1m_in": 0.12,
338 "cost_per_1m_out": 0.2,
339 "cost_per_1m_in_cached": 0,
340 "cost_per_1m_out_cached": 0,
341 "context_window": 198000,
342 "default_max_tokens": 32768,
343 "can_reason": false,
344 "supports_attachments": true,
345 "options": {}
346 },
347 {
348 "id": "grok-41-fast",
349 "name": "Grok 4.1 Fast",
350 "cost_per_1m_in": 0.25,
351 "cost_per_1m_out": 0.625,
352 "cost_per_1m_in_cached": 0,
353 "cost_per_1m_out_cached": 0,
354 "context_window": 1000000,
355 "default_max_tokens": 32768,
356 "can_reason": true,
357 "reasoning_levels": [
358 "low",
359 "medium",
360 "high"
361 ],
362 "default_reasoning_effort": "medium",
363 "supports_attachments": true,
364 "options": {}
365 },
366 {
367 "id": "grok-code-fast-1",
368 "name": "Grok Code Fast 1",
369 "cost_per_1m_in": 0.25,
370 "cost_per_1m_out": 1.87,
371 "cost_per_1m_in_cached": 0,
372 "cost_per_1m_out_cached": 0,
373 "context_window": 256000,
374 "default_max_tokens": 32768,
375 "can_reason": true,
376 "reasoning_levels": [
377 "low",
378 "medium",
379 "high"
380 ],
381 "default_reasoning_effort": "medium",
382 "supports_attachments": false,
383 "options": {}
384 },
385 {
386 "id": "kimi-k2-thinking",
387 "name": "Kimi K2 Thinking",
388 "cost_per_1m_in": 0.75,
389 "cost_per_1m_out": 3.2,
390 "cost_per_1m_in_cached": 0,
391 "cost_per_1m_out_cached": 0,
392 "context_window": 256000,
393 "default_max_tokens": 32768,
394 "can_reason": true,
395 "reasoning_levels": [
396 "low",
397 "medium",
398 "high"
399 ],
400 "default_reasoning_effort": "medium",
401 "supports_attachments": false,
402 "options": {}
403 },
404 {
405 "id": "kimi-k2-5",
406 "name": "Kimi K2.5",
407 "cost_per_1m_in": 0.75,
408 "cost_per_1m_out": 3.75,
409 "cost_per_1m_in_cached": 0,
410 "cost_per_1m_out_cached": 0,
411 "context_window": 256000,
412 "default_max_tokens": 32768,
413 "can_reason": true,
414 "reasoning_levels": [
415 "low",
416 "medium",
417 "high"
418 ],
419 "default_reasoning_effort": "medium",
420 "supports_attachments": true,
421 "options": {}
422 },
423 {
424 "id": "llama-3.2-3b",
425 "name": "Llama 3.2 3B",
426 "cost_per_1m_in": 0.15,
427 "cost_per_1m_out": 0.6,
428 "cost_per_1m_in_cached": 0,
429 "cost_per_1m_out_cached": 0,
430 "context_window": 128000,
431 "default_max_tokens": 32000,
432 "can_reason": false,
433 "supports_attachments": false,
434 "options": {}
435 },
436 {
437 "id": "llama-3.3-70b",
438 "name": "Llama 3.3 70B",
439 "cost_per_1m_in": 0.7,
440 "cost_per_1m_out": 2.8,
441 "cost_per_1m_in_cached": 0,
442 "cost_per_1m_out_cached": 0,
443 "context_window": 128000,
444 "default_max_tokens": 32000,
445 "can_reason": false,
446 "supports_attachments": false,
447 "options": {}
448 },
449 {
450 "id": "minimax-m21",
451 "name": "MiniMax M2.1",
452 "cost_per_1m_in": 0.4,
453 "cost_per_1m_out": 1.6,
454 "cost_per_1m_in_cached": 0,
455 "cost_per_1m_out_cached": 0,
456 "context_window": 198000,
457 "default_max_tokens": 32768,
458 "can_reason": true,
459 "reasoning_levels": [
460 "low",
461 "medium",
462 "high"
463 ],
464 "default_reasoning_effort": "medium",
465 "supports_attachments": false,
466 "options": {}
467 },
468 {
469 "id": "minimax-m25",
470 "name": "MiniMax M2.5",
471 "cost_per_1m_in": 0.4,
472 "cost_per_1m_out": 1.6,
473 "cost_per_1m_in_cached": 0,
474 "cost_per_1m_out_cached": 0,
475 "context_window": 198000,
476 "default_max_tokens": 32768,
477 "can_reason": true,
478 "reasoning_levels": [
479 "low",
480 "medium",
481 "high"
482 ],
483 "default_reasoning_effort": "medium",
484 "supports_attachments": false,
485 "options": {}
486 },
487 {
488 "id": "nvidia-nemotron-3-nano-30b-a3b",
489 "name": "NVIDIA Nemotron 3 Nano 30B",
490 "cost_per_1m_in": 0.075,
491 "cost_per_1m_out": 0.3,
492 "cost_per_1m_in_cached": 0,
493 "cost_per_1m_out_cached": 0,
494 "context_window": 128000,
495 "default_max_tokens": 32000,
496 "can_reason": false,
497 "supports_attachments": false,
498 "options": {}
499 },
500 {
501 "id": "openai-gpt-oss-120b",
502 "name": "OpenAI GPT OSS 120B",
503 "cost_per_1m_in": 0.07,
504 "cost_per_1m_out": 0.3,
505 "cost_per_1m_in_cached": 0,
506 "cost_per_1m_out_cached": 0,
507 "context_window": 128000,
508 "default_max_tokens": 32000,
509 "can_reason": false,
510 "supports_attachments": false,
511 "options": {}
512 },
513 {
514 "id": "qwen3-235b-a22b-instruct-2507",
515 "name": "Qwen 3 235B A22B Instruct 2507",
516 "cost_per_1m_in": 0.15,
517 "cost_per_1m_out": 0.75,
518 "cost_per_1m_in_cached": 0,
519 "cost_per_1m_out_cached": 0,
520 "context_window": 128000,
521 "default_max_tokens": 32000,
522 "can_reason": false,
523 "supports_attachments": false,
524 "options": {}
525 },
526 {
527 "id": "qwen3-235b-a22b-thinking-2507",
528 "name": "Qwen 3 235B A22B Thinking 2507",
529 "cost_per_1m_in": 0.45,
530 "cost_per_1m_out": 3.5,
531 "cost_per_1m_in_cached": 0,
532 "cost_per_1m_out_cached": 0,
533 "context_window": 128000,
534 "default_max_tokens": 32000,
535 "can_reason": true,
536 "reasoning_levels": [
537 "low",
538 "medium",
539 "high"
540 ],
541 "default_reasoning_effort": "medium",
542 "supports_attachments": false,
543 "options": {}
544 },
545 {
546 "id": "qwen3-coder-480b-a35b-instruct-turbo",
547 "name": "Qwen 3 Coder 480B Turbo",
548 "cost_per_1m_in": 0.35,
549 "cost_per_1m_out": 1.5,
550 "cost_per_1m_in_cached": 0,
551 "cost_per_1m_out_cached": 0,
552 "context_window": 256000,
553 "default_max_tokens": 32768,
554 "can_reason": false,
555 "supports_attachments": false,
556 "options": {}
557 },
558 {
559 "id": "qwen3-coder-480b-a35b-instruct",
560 "name": "Qwen 3 Coder 480b",
561 "cost_per_1m_in": 0.75,
562 "cost_per_1m_out": 3,
563 "cost_per_1m_in_cached": 0,
564 "cost_per_1m_out_cached": 0,
565 "context_window": 256000,
566 "default_max_tokens": 32768,
567 "can_reason": false,
568 "supports_attachments": false,
569 "options": {}
570 },
571 {
572 "id": "qwen3-next-80b",
573 "name": "Qwen 3 Next 80b",
574 "cost_per_1m_in": 0.35,
575 "cost_per_1m_out": 1.9,
576 "cost_per_1m_in_cached": 0,
577 "cost_per_1m_out_cached": 0,
578 "context_window": 256000,
579 "default_max_tokens": 32768,
580 "can_reason": false,
581 "supports_attachments": false,
582 "options": {}
583 },
584 {
585 "id": "qwen3-5-35b-a3b",
586 "name": "Qwen 3.5 35B A3B",
587 "cost_per_1m_in": 0.3125,
588 "cost_per_1m_out": 1.25,
589 "cost_per_1m_in_cached": 0,
590 "cost_per_1m_out_cached": 0,
591 "context_window": 256000,
592 "default_max_tokens": 32768,
593 "can_reason": true,
594 "reasoning_levels": [
595 "low",
596 "medium",
597 "high"
598 ],
599 "default_reasoning_effort": "medium",
600 "supports_attachments": true,
601 "options": {
602 "temperature": 1,
603 "top_p": 0.95
604 }
605 },
606 {
607 "id": "qwen3-vl-235b-a22b",
608 "name": "Qwen3 VL 235B",
609 "cost_per_1m_in": 0.25,
610 "cost_per_1m_out": 1.5,
611 "cost_per_1m_in_cached": 0,
612 "cost_per_1m_out_cached": 0,
613 "context_window": 256000,
614 "default_max_tokens": 32768,
615 "can_reason": false,
616 "supports_attachments": true,
617 "options": {}
618 },
619 {
620 "id": "mistral-31-24b",
621 "name": "Venice Medium",
622 "cost_per_1m_in": 0.5,
623 "cost_per_1m_out": 2,
624 "cost_per_1m_in_cached": 0,
625 "cost_per_1m_out_cached": 0,
626 "context_window": 128000,
627 "default_max_tokens": 32000,
628 "can_reason": false,
629 "supports_attachments": true,
630 "options": {}
631 },
632 {
633 "id": "qwen3-4b",
634 "name": "Venice Small",
635 "cost_per_1m_in": 0.05,
636 "cost_per_1m_out": 0.15,
637 "cost_per_1m_in_cached": 0,
638 "cost_per_1m_out_cached": 0,
639 "context_window": 32000,
640 "default_max_tokens": 8000,
641 "can_reason": true,
642 "reasoning_levels": [
643 "low",
644 "medium",
645 "high"
646 ],
647 "default_reasoning_effort": "medium",
648 "supports_attachments": false,
649 "options": {}
650 }
651 ]
652}