1{
2 "name": "Chutes",
3 "id": "chutes",
4 "type": "openai-compat",
5 "api_key": "$CHUTES_API_KEY",
6 "api_endpoint": "https://llm.chutes.ai/v1",
7 "default_large_model_id": "Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8",
8 "default_small_model_id": "Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8",
9 "models": [
10 {
11 "id": "Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8",
12 "name": "Qwen3 Coder 480B A35B Instruct (FP8)",
13 "cost_per_1m_in": 0.2,
14 "cost_per_1m_out": 0.8,
15 "context_window": 262000,
16 "default_max_tokens": 32768,
17 "can_reason": true,
18 "reasoning_levels": [
19 "low",
20 "medium",
21 "high"
22 ],
23 "default_reasoning_efforts": "medium",
24 "supports_attachments": true
25 },
26 {
27 "id": "zai-org/GLM-4.5-FP8",
28 "name": "GLM 4.5 FP8",
29 "cost_per_1m_in": 0.0,
30 "cost_per_1m_out": 0.0,
31 "context_window": 98000,
32 "default_max_tokens": 32768,
33 "can_reason": true,
34 "reasoning_levels": [
35 "low",
36 "medium",
37 "high"
38 ],
39 "default_reasoning_efforts": "medium",
40 "supports_attachments": true
41 },
42 {
43 "id": "moonshotai/Kimi-K2-Instruct-75k",
44 "name": "Kimi K2 Instruct",
45 "cost_per_1m_in": 0.15,
46 "cost_per_1m_out": 0.59,
47 "context_window": 75000,
48 "default_max_tokens": 32768,
49 "can_reason": true,
50 "reasoning_levels": [
51 "low",
52 "medium",
53 "high"
54 ],
55 "default_reasoning_efforts": "medium",
56 "supports_attachments": true
57 },
58 {
59 "id": "deepseek-ai/DeepSeek-R1-0528",
60 "name": "DeepSeek R1 0528",
61 "cost_per_1m_in": 0.18,
62 "cost_per_1m_out": 0.72,
63 "context_window": 75000,
64 "default_max_tokens": 32768,
65 "can_reason": true,
66 "reasoning_levels": [
67 "low",
68 "medium",
69 "high"
70 ],
71 "default_reasoning_efforts": "medium",
72 "supports_attachments": true
73 },
74 {
75 "id": "deepseek-ai/DeepSeek-R1-0528-Qwen3-8B",
76 "name": "DeepSeek R1 0528 Qwen3 8B",
77 "cost_per_1m_in": 0.02,
78 "cost_per_1m_out": 0.07,
79 "context_window": 32768,
80 "default_max_tokens": 8192,
81 "can_reason": false,
82 "has_reasoning_efforts": false,
83 "supports_attachments": true
84 },
85 {
86 "id": "deepseek-ai/DeepSeek-R1-Distill-Llama-70B",
87 "name": "DeepSeek R1 Distill Llama 70B",
88 "cost_per_1m_in": 0.03,
89 "cost_per_1m_out": 0.14,
90 "context_window": 65536,
91 "default_max_tokens": 8192,
92 "can_reason": false,
93 "has_reasoning_efforts": false,
94 "supports_attachments": true
95 },
96 {
97 "id": "tngtech/DeepSeek-R1T-Chimera",
98 "name": "DeepSeek R1T Chimera",
99 "cost_per_1m_in": 0.18,
100 "cost_per_1m_out": 0.72,
101 "context_window": 131072,
102 "default_max_tokens": 32768,
103 "can_reason": true,
104 "reasoning_levels": [
105 "low",
106 "medium",
107 "high"
108 ],
109 "default_reasoning_efforts": "medium",
110 "supports_attachments": true
111 },
112 {
113 "id": "tngtech/DeepSeek-TNG-R1T2-Chimera",
114 "name": "DeepSeek TNG R1T2 Chimera",
115 "cost_per_1m_in": 0.2,
116 "cost_per_1m_out": 0.8,
117 "context_window": 262144,
118 "default_max_tokens": 65536,
119 "can_reason": true,
120 "reasoning_levels": [
121 "low",
122 "medium",
123 "high"
124 ],
125 "default_reasoning_efforts": "medium",
126 "supports_attachments": true
127 },
128 {
129 "id": "deepseek-ai/DeepSeek-V3-0324",
130 "name": "DeepSeek V3 0324",
131 "cost_per_1m_in": 0.18,
132 "cost_per_1m_out": 0.72,
133 "context_window": 75000,
134 "default_max_tokens": 32768,
135 "can_reason": true,
136 "reasoning_levels": [
137 "low",
138 "medium",
139 "high"
140 ],
141 "default_reasoning_efforts": "medium",
142 "supports_attachments": true
143 },
144 {
145 "id": "chutesai/Devstral-Small-2505",
146 "name": "Devstral Small 2505",
147 "cost_per_1m_in": 0.02,
148 "cost_per_1m_out": 0.08,
149 "context_window": 32768,
150 "default_max_tokens": 8192,
151 "can_reason": false,
152 "has_reasoning_efforts": false,
153 "supports_attachments": true
154 },
155 {
156 "id": "zai-org/GLM-4.5-Air",
157 "name": "GLM 4.5 Air",
158 "cost_per_1m_in": 0.0,
159 "cost_per_1m_out": 0.0,
160 "context_window": 131072,
161 "default_max_tokens": 32768,
162 "can_reason": true,
163 "reasoning_levels": [
164 "low",
165 "medium",
166 "high"
167 ],
168 "default_reasoning_efforts": "medium",
169 "supports_attachments": true
170 },
171 {
172 "id": "openai/gpt-oss-120b",
173 "name": "GPT OSS 120B",
174 "cost_per_1m_in": 0.1,
175 "cost_per_1m_out": 0.41,
176 "context_window": 131072,
177 "default_max_tokens": 32768,
178 "can_reason": true,
179 "reasoning_levels": [
180 "low",
181 "medium",
182 "high"
183 ],
184 "default_reasoning_efforts": "medium",
185 "supports_attachments": true
186 },
187 {
188 "id": "chutesai/Mistral-Small-3.2-24B-Instruct-2506",
189 "name": "Mistral Small 3.2 24B Instruct 2506",
190 "cost_per_1m_in": 0.02,
191 "cost_per_1m_out": 0.08,
192 "context_window": 32768,
193 "default_max_tokens": 8192,
194 "can_reason": false,
195 "has_reasoning_efforts": false,
196 "supports_attachments": true
197 },
198 {
199 "id": "Qwen/Qwen3-235B-A22B-Instruct-2507",
200 "name": "Qwen3 235B A22B Instruct 2507",
201 "cost_per_1m_in": 0.08,
202 "cost_per_1m_out": 0.31,
203 "context_window": 32768,
204 "default_max_tokens": 8192,
205 "can_reason": false,
206 "has_reasoning_efforts": false,
207 "supports_attachments": true
208 },
209 {
210 "id": "Qwen/Qwen3-30B-A3B",
211 "name": "Qwen3 30B A3B",
212 "cost_per_1m_in": 0.02,
213 "cost_per_1m_out": 0.08,
214 "context_window": 32768,
215 "default_max_tokens": 8192,
216 "can_reason": false,
217 "has_reasoning_efforts": false,
218 "supports_attachments": true
219 },
220 {
221 "id": "Qwen/Qwen3-235B-A22B-Thinking-2507",
222 "name": "Qwen3 235B A22B Thinking 2507",
223 "cost_per_1m_in": 0.08,
224 "cost_per_1m_out": 0.31,
225 "context_window": 32768,
226 "default_max_tokens": 8192,
227 "can_reason": true,
228 "reasoning_levels": [
229 "low",
230 "medium",
231 "high"
232 ],
233 "default_reasoning_efforts": "medium",
234 "supports_attachments": true
235 },
236 {
237 "id": "deepseek-ai/DeepSeek-V3.1",
238 "name": "DeepSeek V3.1",
239 "cost_per_1m_in": 0.2,
240 "cost_per_1m_out": 0.8,
241 "context_window": 163840,
242 "default_max_tokens": 32768,
243 "can_reason": false,
244 "has_reasoning_efforts": false,
245 "supports_attachments": true
246 },
247 {
248 "id": "deepseek-ai/DeepSeek-V3.1:THINKING",
249 "name": "DeepSeek V3.1 Reasoning",
250 "cost_per_1m_in": 0.2,
251 "cost_per_1m_out": 0.8,
252 "context_window": 163840,
253 "default_max_tokens": 32768,
254 "can_reason": true,
255 "reasoning_levels": [
256 "low",
257 "medium",
258 "high"
259 ],
260 "default_reasoning_efforts": "medium",
261 "supports_attachments": true
262 },
263 {
264 "id": "Qwen/Qwen3-30B-A3B-Instruct-2507",
265 "name": "Qwen3 30B A3B Instruct 2507",
266 "cost_per_1m_in": 0.05,
267 "cost_per_1m_out": 0.2,
268 "context_window": 262144,
269 "default_max_tokens": 32768,
270 "can_reason": false,
271 "has_reasoning_efforts": false,
272 "supports_attachments": true
273 },
274 {
275 "id": "Qwen/Qwen3-Coder-30B-A3B-Instruct",
276 "name": "Qwen3 Coder 30B A3B Instruct",
277 "cost_per_1m_in": 0.0,
278 "cost_per_1m_out": 0.0,
279 "context_window": 262144,
280 "default_max_tokens": 32768,
281 "can_reason": false,
282 "has_reasoning_efforts": false,
283 "supports_attachments": true
284 }
285 ]
286}