cherry-studio/packages/catalog/data/overrides/deepinfra.json
2025-11-24 08:55:12 +08:00

1548 lines
40 KiB
JSON

{
"version": "2025.11.24",
"overrides": [
{
"providerId": "deepinfra",
"modelId": "Gryphe/MythoMax-L2-13b",
"disabled": false,
"reason": "Provider-specific implementation of Gryphe/MythoMax-L2-13b",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 4096
},
"pricing": {
"input": {
"perMillionTokens": 0.08,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.09,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "NousResearch/Hermes-3-Llama-3.1-405B",
"disabled": false,
"reason": "Provider-specific implementation of NousResearch/Hermes-3-Llama-3.1-405B",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 1,
"currency": "USD"
},
"output": {
"perMillionTokens": 1,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "NousResearch/Hermes-3-Llama-3.1-70B",
"disabled": false,
"reason": "Provider-specific implementation of NousResearch/Hermes-3-Llama-3.1-70B",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.3,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.3,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "Qwen/QwQ-32B",
"disabled": false,
"reason": "Provider-specific implementation of Qwen/QwQ-32B",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.15,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.4,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "Qwen/Qwen2.5-72B-Instruct",
"disabled": false,
"reason": "Provider-specific implementation of Qwen/Qwen2.5-72B-Instruct",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 32768,
"maxOutputTokens": 32768
},
"pricing": {
"input": {
"perMillionTokens": 0.12,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.39,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "Qwen/Qwen2.5-7B-Instruct",
"disabled": false,
"reason": "Provider-specific implementation of Qwen/Qwen2.5-7B-Instruct",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 32768,
"maxOutputTokens": 32768
},
"pricing": {
"input": {
"perMillionTokens": 0.04,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.1,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "Qwen/Qwen2.5-VL-32B-Instruct",
"disabled": false,
"reason": "Provider-specific implementation of Qwen/Qwen2.5-VL-32B-Instruct",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"capabilities": {
"add": [
"IMAGE_RECOGNITION"
]
},
"limits": {
"maxOutputTokens": 128000
},
"pricing": {
"input": {
"perMillionTokens": 0.2,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.6,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "Qwen/Qwen3-14B",
"disabled": false,
"reason": "Provider-specific implementation of Qwen/Qwen3-14B",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 40960,
"maxOutputTokens": 40960
},
"pricing": {
"input": {
"perMillionTokens": 0.06,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.24,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "Qwen/Qwen3-235B-A22B",
"disabled": false,
"reason": "Provider-specific implementation of Qwen/Qwen3-235B-A22B",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 40960,
"maxOutputTokens": 40960
},
"pricing": {
"input": {
"perMillionTokens": 0.18,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.54,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "Qwen/Qwen3-235B-A22B-Instruct-2507",
"disabled": false,
"reason": "Provider-specific implementation of Qwen/Qwen3-235B-A22B-Instruct-2507",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 262144,
"maxOutputTokens": 262144
},
"pricing": {
"input": {
"perMillionTokens": 0.09,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.6,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "Qwen/Qwen3-235B-A22B-Thinking-2507",
"disabled": false,
"reason": "Provider-specific implementation of Qwen/Qwen3-235B-A22B-Thinking-2507",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 262144,
"maxOutputTokens": 262144
},
"pricing": {
"input": {
"perMillionTokens": 0.3,
"currency": "USD"
},
"output": {
"perMillionTokens": 2.9,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "Qwen/Qwen3-30B-A3B",
"disabled": false,
"reason": "Provider-specific implementation of Qwen/Qwen3-30B-A3B",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 40960,
"maxOutputTokens": 40960
},
"pricing": {
"input": {
"perMillionTokens": 0.08,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.29,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "Qwen/Qwen3-32B",
"disabled": false,
"reason": "Provider-specific implementation of Qwen/Qwen3-32B",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 40960,
"maxOutputTokens": 40960
},
"pricing": {
"input": {
"perMillionTokens": 0.1,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.28,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "Qwen/Qwen3-Coder-480B-A35B-Instruct",
"disabled": false,
"reason": "Provider-specific implementation of Qwen/Qwen3-Coder-480B-A35B-Instruct",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 262144,
"maxOutputTokens": 262144
},
"pricing": {
"input": {
"perMillionTokens": 0.4,
"currency": "USD"
},
"output": {
"perMillionTokens": 1.6,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "Qwen/Qwen3-Coder-480B-A35B-Instruct-Turbo",
"disabled": false,
"reason": "Provider-specific implementation of Qwen/Qwen3-Coder-480B-A35B-Instruct-Turbo",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 262144,
"maxOutputTokens": 262144
},
"pricing": {
"input": {
"perMillionTokens": 0.29,
"currency": "USD"
},
"output": {
"perMillionTokens": 1.2,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "Qwen/Qwen3-Next-80B-A3B-Instruct",
"disabled": false,
"reason": "Provider-specific implementation of Qwen/Qwen3-Next-80B-A3B-Instruct",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 262144,
"maxOutputTokens": 262144
},
"pricing": {
"input": {
"perMillionTokens": 0.14,
"currency": "USD"
},
"output": {
"perMillionTokens": 1.4,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "Qwen/Qwen3-Next-80B-A3B-Thinking",
"disabled": false,
"reason": "Provider-specific implementation of Qwen/Qwen3-Next-80B-A3B-Thinking",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 262144,
"maxOutputTokens": 262144
},
"pricing": {
"input": {
"perMillionTokens": 0.14,
"currency": "USD"
},
"output": {
"perMillionTokens": 1.4,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "Sao10K/L3-8B-Lunaris-v1-Turbo",
"disabled": false,
"reason": "Provider-specific implementation of Sao10K/L3-8B-Lunaris-v1-Turbo",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 8192,
"maxOutputTokens": 8192
},
"pricing": {
"input": {
"perMillionTokens": 0.04,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.05,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "Sao10K/L3.1-70B-Euryale-v2.2",
"disabled": false,
"reason": "Provider-specific implementation of Sao10K/L3.1-70B-Euryale-v2.2",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.65,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.75,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "Sao10K/L3.3-70B-Euryale-v2.3",
"disabled": false,
"reason": "Provider-specific implementation of Sao10K/L3.3-70B-Euryale-v2.3",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.65,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.75,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "allenai/olmOCR-7B-0725-FP8",
"disabled": false,
"reason": "Provider-specific implementation of allenai/olmOCR-7B-0725-FP8",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 16384,
"maxOutputTokens": 16384
},
"pricing": {
"input": {
"perMillionTokens": 0.27,
"currency": "USD"
},
"output": {
"perMillionTokens": 1.5,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "anthropic/claude-3-7-sonnet-latest",
"disabled": false,
"reason": "Provider-specific implementation of anthropic/claude-3-7-sonnet-latest",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 200000,
"maxOutputTokens": 200000
},
"pricing": {
"input": {
"perMillionTokens": 3.3,
"currency": "USD"
},
"output": {
"perMillionTokens": 16.5,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "anthropic/claude-4-opus",
"disabled": false,
"reason": "Provider-specific implementation of anthropic/claude-4-opus",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 200000,
"maxOutputTokens": 200000
},
"pricing": {
"input": {
"perMillionTokens": 16.5,
"currency": "USD"
},
"output": {
"perMillionTokens": 82.5,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "anthropic/claude-4-sonnet",
"disabled": false,
"reason": "Provider-specific implementation of anthropic/claude-4-sonnet",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 200000,
"maxOutputTokens": 200000
},
"pricing": {
"input": {
"perMillionTokens": 3.3,
"currency": "USD"
},
"output": {
"perMillionTokens": 16.5,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "deepseek-ai/DeepSeek-R1",
"disabled": false,
"reason": "Provider-specific implementation of deepseek-ai/DeepSeek-R1",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 163840,
"maxOutputTokens": 163840
},
"pricing": {
"input": {
"perMillionTokens": 0.7,
"currency": "USD"
},
"output": {
"perMillionTokens": 2.4,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "deepseek-ai/DeepSeek-R1-0528",
"disabled": false,
"reason": "Provider-specific implementation of deepseek-ai/DeepSeek-R1-0528",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 163840,
"maxOutputTokens": 163840
},
"pricing": {
"input": {
"perMillionTokens": 0.5,
"currency": "USD"
},
"output": {
"perMillionTokens": 2.15,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "deepseek-ai/DeepSeek-R1-0528-Turbo",
"disabled": false,
"reason": "Provider-specific implementation of deepseek-ai/DeepSeek-R1-0528-Turbo",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 32768,
"maxOutputTokens": 32768
},
"pricing": {
"input": {
"perMillionTokens": 1,
"currency": "USD"
},
"output": {
"perMillionTokens": 3,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "deepseek-ai/DeepSeek-R1-Distill-Llama-70B",
"disabled": false,
"reason": "Provider-specific implementation of deepseek-ai/DeepSeek-R1-Distill-Llama-70B",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.2,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.6,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B",
"disabled": false,
"reason": "Provider-specific implementation of deepseek-ai/DeepSeek-R1-Distill-Qwen-32B",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.27,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.27,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "deepseek-ai/DeepSeek-R1-Turbo",
"disabled": false,
"reason": "Provider-specific implementation of deepseek-ai/DeepSeek-R1-Turbo",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 40960,
"maxOutputTokens": 40960
},
"pricing": {
"input": {
"perMillionTokens": 1,
"currency": "USD"
},
"output": {
"perMillionTokens": 3,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "deepseek-ai/DeepSeek-V3",
"disabled": false,
"reason": "Provider-specific implementation of deepseek-ai/DeepSeek-V3",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 163840,
"maxOutputTokens": 163840
},
"pricing": {
"input": {
"perMillionTokens": 0.38,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.89,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "deepseek-ai/DeepSeek-V3-0324",
"disabled": false,
"reason": "Provider-specific implementation of deepseek-ai/DeepSeek-V3-0324",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 163840,
"maxOutputTokens": 163840
},
"pricing": {
"input": {
"perMillionTokens": 0.25,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.88,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "deepseek-ai/DeepSeek-V3.1",
"disabled": false,
"reason": "Provider-specific implementation of deepseek-ai/DeepSeek-V3.1",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 163840,
"maxOutputTokens": 163840
},
"pricing": {
"input": {
"perMillionTokens": 0.27,
"currency": "USD"
},
"output": {
"perMillionTokens": 1,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "deepseek-ai/DeepSeek-V3.1-Terminus",
"disabled": false,
"reason": "Provider-specific implementation of deepseek-ai/DeepSeek-V3.1-Terminus",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 163840,
"maxOutputTokens": 163840
},
"pricing": {
"input": {
"perMillionTokens": 0.27,
"currency": "USD"
},
"output": {
"perMillionTokens": 1,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "google/gemini-2.0-flash-001",
"disabled": false,
"reason": "Provider-specific implementation of google/gemini-2.0-flash-001",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 1000000,
"maxOutputTokens": 1000000
},
"pricing": {
"input": {
"perMillionTokens": 0.1,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.4,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "google/gemini-2.5-flash",
"disabled": false,
"reason": "Provider-specific implementation of google/gemini-2.5-flash",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 1000000,
"maxOutputTokens": 1000000
},
"pricing": {
"input": {
"perMillionTokens": 0.3,
"currency": "USD"
},
"output": {
"perMillionTokens": 2.5,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "google/gemini-2.5-pro",
"disabled": false,
"reason": "Provider-specific implementation of google/gemini-2.5-pro",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 1000000,
"maxOutputTokens": 1000000
},
"pricing": {
"input": {
"perMillionTokens": 1.25,
"currency": "USD"
},
"output": {
"perMillionTokens": 10,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "google/gemma-3-12b-it",
"disabled": false,
"reason": "Provider-specific implementation of google/gemma-3-12b-it",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.05,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.1,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "google/gemma-3-27b-it",
"disabled": false,
"reason": "Provider-specific implementation of google/gemma-3-27b-it",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.09,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.16,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "google/gemma-3-4b-it",
"disabled": false,
"reason": "Provider-specific implementation of google/gemma-3-4b-it",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.04,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.08,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "meta-llama/Llama-3.2-11B-Vision-Instruct",
"disabled": false,
"reason": "Provider-specific implementation of meta-llama/Llama-3.2-11B-Vision-Instruct",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.049,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.049,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "meta-llama/Llama-3.2-3B-Instruct",
"disabled": false,
"reason": "Provider-specific implementation of meta-llama/Llama-3.2-3B-Instruct",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.02,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.02,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "meta-llama/Llama-3.3-70B-Instruct",
"disabled": false,
"reason": "Provider-specific implementation of meta-llama/Llama-3.3-70B-Instruct",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.23,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.4,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "meta-llama/Llama-3.3-70B-Instruct-Turbo",
"disabled": false,
"reason": "Provider-specific implementation of meta-llama/Llama-3.3-70B-Instruct-Turbo",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.13,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.39,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8",
"disabled": false,
"reason": "Provider-specific implementation of meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 1048576,
"maxOutputTokens": 1048576
},
"pricing": {
"input": {
"perMillionTokens": 0.15,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.6,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "meta-llama/Llama-4-Scout-17B-16E-Instruct",
"disabled": false,
"reason": "Provider-specific implementation of meta-llama/Llama-4-Scout-17B-16E-Instruct",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 327680,
"maxOutputTokens": 327680
},
"pricing": {
"input": {
"perMillionTokens": 0.08,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.3,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "meta-llama/Llama-Guard-3-8B",
"disabled": false,
"reason": "Provider-specific implementation of meta-llama/Llama-Guard-3-8B",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.055,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.055,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "meta-llama/Llama-Guard-4-12B",
"disabled": false,
"reason": "Provider-specific implementation of meta-llama/Llama-Guard-4-12B",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 163840,
"maxOutputTokens": 163840
},
"pricing": {
"input": {
"perMillionTokens": 0.18,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.18,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "meta-llama/Meta-Llama-3-8B-Instruct",
"disabled": false,
"reason": "Provider-specific implementation of meta-llama/Meta-Llama-3-8B-Instruct",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 8192,
"maxOutputTokens": 8192
},
"pricing": {
"input": {
"perMillionTokens": 0.03,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.06,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "meta-llama/Meta-Llama-3.1-70B-Instruct",
"disabled": false,
"reason": "Provider-specific implementation of meta-llama/Meta-Llama-3.1-70B-Instruct",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.4,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.4,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo",
"disabled": false,
"reason": "Provider-specific implementation of meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.1,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.28,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"disabled": false,
"reason": "Provider-specific implementation of meta-llama/Meta-Llama-3.1-8B-Instruct",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.03,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.05,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo",
"disabled": false,
"reason": "Provider-specific implementation of meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.02,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.03,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "microsoft/WizardLM-2-8x22B",
"disabled": false,
"reason": "Provider-specific implementation of microsoft/WizardLM-2-8x22B",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 65536,
"maxOutputTokens": 65536
},
"pricing": {
"input": {
"perMillionTokens": 0.48,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.48,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "microsoft/phi-4",
"disabled": false,
"reason": "Provider-specific implementation of microsoft/phi-4",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 16384,
"maxOutputTokens": 16384
},
"pricing": {
"input": {
"perMillionTokens": 0.07,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.14,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "mistralai/Mistral-Nemo-Instruct-2407",
"disabled": false,
"reason": "Provider-specific implementation of mistralai/Mistral-Nemo-Instruct-2407",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.02,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.04,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "mistralai/Mistral-Small-24B-Instruct-2501",
"disabled": false,
"reason": "Provider-specific implementation of mistralai/Mistral-Small-24B-Instruct-2501",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 32768,
"maxOutputTokens": 32768
},
"pricing": {
"input": {
"perMillionTokens": 0.05,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.08,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "mistralai/Mistral-Small-3.2-24B-Instruct-2506",
"disabled": false,
"reason": "Provider-specific implementation of mistralai/Mistral-Small-3.2-24B-Instruct-2506",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"maxOutputTokens": 128000
},
"pricing": {
"input": {
"perMillionTokens": 0.075,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.2,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "mistralai/Mixtral-8x7B-Instruct-v0.1",
"disabled": false,
"reason": "Provider-specific implementation of mistralai/Mixtral-8x7B-Instruct-v0.1",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 32768,
"maxOutputTokens": 32768
},
"pricing": {
"input": {
"perMillionTokens": 0.4,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.4,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "moonshotai/Kimi-K2-Instruct",
"disabled": false,
"reason": "Provider-specific implementation of moonshotai/Kimi-K2-Instruct",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.5,
"currency": "USD"
},
"output": {
"perMillionTokens": 2,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "moonshotai/Kimi-K2-Instruct-0905",
"disabled": false,
"reason": "Provider-specific implementation of moonshotai/Kimi-K2-Instruct-0905",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 262144,
"maxOutputTokens": 262144
},
"pricing": {
"input": {
"perMillionTokens": 0.5,
"currency": "USD"
},
"output": {
"perMillionTokens": 2,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "nvidia/Llama-3.1-Nemotron-70B-Instruct",
"disabled": false,
"reason": "Provider-specific implementation of nvidia/Llama-3.1-Nemotron-70B-Instruct",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.6,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.6,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "nvidia/Llama-3.3-Nemotron-Super-49B-v1.5",
"disabled": false,
"reason": "Provider-specific implementation of nvidia/Llama-3.3-Nemotron-Super-49B-v1.5",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.1,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.4,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "nvidia/NVIDIA-Nemotron-Nano-9B-v2",
"disabled": false,
"reason": "Provider-specific implementation of nvidia/NVIDIA-Nemotron-Nano-9B-v2",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.04,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.16,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "openai/gpt-oss-120b",
"disabled": false,
"reason": "Provider-specific implementation of openai/gpt-oss-120b",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.05,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.45,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "openai/gpt-oss-20b",
"disabled": false,
"reason": "Provider-specific implementation of openai/gpt-oss-20b",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.04,
"currency": "USD"
},
"output": {
"perMillionTokens": 0.15,
"currency": "USD"
}
}
},
{
"providerId": "deepinfra",
"modelId": "zai-org/GLM-4.5",
"disabled": false,
"reason": "Provider-specific implementation of zai-org/GLM-4.5",
"lastUpdated": "2025-11-23",
"updatedBy": "migration-tool",
"priority": 100,
"limits": {
"contextWindow": 131072,
"maxOutputTokens": 131072
},
"pricing": {
"input": {
"perMillionTokens": 0.4,
"currency": "USD"
},
"output": {
"perMillionTokens": 1.6,
"currency": "USD"
}
}
}
]
}