1{
2 "name": "Cerebras",
3 "id": "cerebras",
4 "type": "openai-compat",
5 "api_key": "$CEREBRAS_API_KEY",
6 "api_endpoint": "https://api.cerebras.ai/v1",
7 "default_large_model_id": "qwen-3-coder-480b",
8 "default_small_model_id": "qwen-3-32b",
9 "models": [
10 {
11 "id": "llama-4-scout-17b-16e-instruct",
12 "name": "Llama 4 Scout",
13 "cost_per_1m_in": 0.65,
14 "cost_per_1m_out": 0.85,
15 "context_window": 32768,
16 "default_max_tokens": 4000,
17 "can_reason": false,
18 "supports_attachments": false
19 },
20 {
21 "id": "llama3.1-8b",
22 "name": "Llama 3.1 8B",
23 "cost_per_1m_in": 0.1,
24 "cost_per_1m_out": 0.1,
25 "context_window": 32768,
26 "default_max_tokens": 4000,
27 "can_reason": false,
28 "supports_attachments": false
29 },
30 {
31 "id": "llama-3.3-70b",
32 "name": "Llama 3.3 70B",
33 "cost_per_1m_in": 0.85,
34 "cost_per_1m_out": 1.2,
35 "context_window": 128000,
36 "default_max_tokens": 4000,
37 "can_reason": false,
38 "supports_attachments": false
39 },
40 {
41 "id": "gpt-oss-120b",
42 "name": "gpt-oss-120b",
43 "cost_per_1m_in": 0.4,
44 "cost_per_1m_out": 0.8,
45 "context_window": 128000,
46 "default_max_tokens": 65536,
47 "can_reason": true,
48 "reasoning_levels": [
49 "low",
50 "medium",
51 "high"
52 ],
53 "default_reasoning_efforts": "medium",
54 "supports_attachments": false
55 },
56 {
57 "id": "qwen-3-32b",
58 "name": "Qwen 3 32B",
59 "cost_per_1m_in": 0.4,
60 "cost_per_1m_out": 0.8,
61 "context_window": 128000,
62 "default_max_tokens": 32768,
63 "can_reason": false,
64 "supports_attachments": false
65 },
66 {
67 "id": "llama-4-maverick-17b-128e-instruct",
68 "name": "Llama 4 Maverick",
69 "cost_per_1m_in": 0.2,
70 "cost_per_1m_out": 0.6,
71 "context_window": 32768,
72 "default_max_tokens": 4000,
73 "can_reason": false,
74 "supports_attachments": false
75 },
76 {
77 "id": "qwen-3-235b-a22b-instruct-2507",
78 "name": "Qwen 3 235B Instruct",
79 "cost_per_1m_in": 0.6,
80 "cost_per_1m_out": 1.2,
81 "context_window": 131072,
82 "default_max_tokens": 16384,
83 "can_reason": true,
84 "reasoning_levels": [
85 "low",
86 "medium",
87 "high"
88 ],
89 "default_reasoning_efforts": "medium",
90 "supports_attachments": false
91 },
92 {
93 "id": "qwen-3-235b-a22b-thinking-2507",
94 "name": "Qwen 3 235B Thinking",
95 "cost_per_1m_in": 0.6,
96 "cost_per_1m_out": 1.2,
97 "context_window": 128000,
98 "default_max_tokens": 32768,
99 "can_reason": true,
100 "reasoning_levels": [
101 "low",
102 "medium",
103 "high"
104 ],
105 "default_reasoning_efforts": "medium",
106 "supports_attachments": false
107 },
108 {
109 "id": "qwen-3-coder-480b",
110 "name": "Qwen 3 480B Coder",
111 "cost_per_1m_in": 2.0,
112 "cost_per_1m_out": 2.0,
113 "context_window": 131072,
114 "default_max_tokens": 65536,
115 "can_reason": true,
116 "reasoning_levels": [
117 "low",
118 "medium",
119 "high"
120 ],
121 "default_reasoning_efforts": "medium",
122 "supports_attachments": false
123 }
124 ]
125}