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": "gpt-oss-120b",
8 "default_small_model_id": "qwen-3-32b",
9 "models": [
10 {
11 "id": "zai-glm-4.6",
12 "name": "Z.AI GLM 4.6",
13 "cost_per_1m_in": 0,
14 "cost_per_1m_out": 0,
15 "context_window": 128000,
16 "default_max_tokens": 40000,
17 "can_reason": true,
18 "supports_attachments": false
19 },
20 {
21 "id": "llama-4-scout-17b-16e-instruct",
22 "name": "Llama 4 Scout",
23 "cost_per_1m_in": 0.65,
24 "cost_per_1m_out": 0.85,
25 "context_window": 32768,
26 "default_max_tokens": 4000,
27 "can_reason": false,
28 "supports_attachments": false
29 },
30 {
31 "id": "llama3.1-8b",
32 "name": "Llama 3.1 8B",
33 "cost_per_1m_in": 0.1,
34 "cost_per_1m_out": 0.1,
35 "context_window": 32768,
36 "default_max_tokens": 4000,
37 "can_reason": false,
38 "supports_attachments": false
39 },
40 {
41 "id": "llama-3.3-70b",
42 "name": "Llama 3.3 70B",
43 "cost_per_1m_in": 0.85,
44 "cost_per_1m_out": 1.2,
45 "context_window": 128000,
46 "default_max_tokens": 4000,
47 "can_reason": false,
48 "supports_attachments": false
49 },
50 {
51 "id": "gpt-oss-120b",
52 "name": "gpt-oss-120b",
53 "cost_per_1m_in": 0.4,
54 "cost_per_1m_out": 0.8,
55 "context_window": 128000,
56 "default_max_tokens": 65536,
57 "can_reason": true,
58 "reasoning_levels": [
59 "low",
60 "medium",
61 "high"
62 ],
63 "default_reasoning_efforts": "medium",
64 "supports_attachments": false
65 },
66 {
67 "id": "qwen-3-32b",
68 "name": "Qwen 3 32B",
69 "cost_per_1m_in": 0.4,
70 "cost_per_1m_out": 0.8,
71 "context_window": 128000,
72 "default_max_tokens": 32768,
73 "can_reason": false,
74 "supports_attachments": false
75 },
76 {
77 "id": "llama-4-maverick-17b-128e-instruct",
78 "name": "Llama 4 Maverick",
79 "cost_per_1m_in": 0.2,
80 "cost_per_1m_out": 0.6,
81 "context_window": 32768,
82 "default_max_tokens": 4000,
83 "can_reason": false,
84 "supports_attachments": false
85 },
86 {
87 "id": "qwen-3-235b-a22b-instruct-2507",
88 "name": "Qwen 3 235B Instruct",
89 "cost_per_1m_in": 0.6,
90 "cost_per_1m_out": 1.2,
91 "context_window": 131072,
92 "default_max_tokens": 16384,
93 "can_reason": true,
94 "reasoning_levels": [
95 "low",
96 "medium",
97 "high"
98 ],
99 "default_reasoning_efforts": "medium",
100 "supports_attachments": false
101 },
102 {
103 "id": "qwen-3-235b-a22b-thinking-2507",
104 "name": "Qwen 3 235B Thinking",
105 "cost_per_1m_in": 0.6,
106 "cost_per_1m_out": 1.2,
107 "context_window": 128000,
108 "default_max_tokens": 32768,
109 "can_reason": true,
110 "reasoning_levels": [
111 "low",
112 "medium",
113 "high"
114 ],
115 "default_reasoning_efforts": "medium",
116 "supports_attachments": false
117 },
118 {
119 "id": "qwen-3-coder-480b",
120 "name": "Qwen 3 480B Coder",
121 "cost_per_1m_in": 2.0,
122 "cost_per_1m_out": 2.0,
123 "context_window": 131072,
124 "default_max_tokens": 65536,
125 "can_reason": true,
126 "reasoning_levels": [
127 "low",
128 "medium",
129 "high"
130 ],
131 "default_reasoning_efforts": "medium",
132 "supports_attachments": false
133 }
134 ]
135}