diff --git a/packages/proxy/schema/index.ts b/packages/proxy/schema/index.ts index 735bc599..84b2c85f 100644 --- a/packages/proxy/schema/index.ts +++ b/packages/proxy/schema/index.ts @@ -1009,6 +1009,7 @@ export const AvailableEndpointTypes: { [name: string]: ModelEndpointType[] } = { "zai-org/GLM-5.1": ["together"], "zai-org/GLM-5.2": ["baseten", "together"], "zai-org/GLM-5.2-Fast": ["baseten"], + "moonshotai/Kimi-K3.1-Pro": ["baseten"], "zai-org/GLM-5.3": ["baseten", "together"], "zai-org/GLM-5.3-Fast": ["baseten"], "zai-org/GLM-5.3-Flash": ["baseten", "together"], @@ -1531,6 +1532,11 @@ export const AvailableEndpointTypes: { [name: string]: ModelEndpointType[] } = { "moonshotai.kimi-k3": ["bedrock"], "us.moonshotai.kimi-k3": ["bedrock"], "global.moonshotai.kimi-k3": ["bedrock"], + "accounts/fireworks/models/kimi-k3p1-pro": ["fireworks"], + "databricks-kimi-k3-1-pro": ["databricks"], + "moonshotai.kimi-k3.1-pro": ["bedrock"], + "us.moonshotai.kimi-k3.1-pro": ["bedrock"], + "global.moonshotai.kimi-k3.1-pro": ["bedrock"], "gemini-3.8-live": ["google"], "gemini-3.8-live-extended-thinking": ["google"], "anthropic.claude-opus-5-5": ["bedrock"], diff --git a/packages/proxy/schema/model_list.json b/packages/proxy/schema/model_list.json index 0153b16e..a7a0a30c 100644 --- a/packages/proxy/schema/model_list.json +++ b/packages/proxy/schema/model_list.json @@ -19373,6 +19373,100 @@ "mistral" ] }, + "moonshotai/Kimi-K3.1-Pro": { + "format": "openai", + "flavor": "chat", + "multimodal": true, + "input_cost_per_mil_tokens": 2, + "output_cost_per_mil_tokens": 10, + "input_cache_read_cost_per_mil_tokens": 0.2, + "displayName": "Kimi K3.1 Pro", + "reasoning": true, + "fallback_models": [ + "accounts/fireworks/models/kimi-k3p1-pro", + "databricks-kimi-k3-1-pro", + "global.moonshotai.kimi-k3.1-pro", + "moonshotai.kimi-k3.1-pro", + "moonshotai/kimi-k3.1-pro", + "us.moonshotai.kimi-k3.1-pro" + ], + "max_input_tokens": 1000000, + "available_providers": [ + "baseten" + ] + }, + "accounts/fireworks/models/kimi-k3p1-pro": { + "format": "openai", + "flavor": "chat", + "multimodal": true, + "input_cost_per_mil_tokens": 2, + "output_cost_per_mil_tokens": 10, + "input_cache_read_cost_per_mil_tokens": 0.2, + "displayName": "Kimi K3.1 Pro", + "reasoning": true, + "max_input_tokens": 1000000, + "available_providers": [ + "fireworks" + ] + }, + "databricks-kimi-k3-1-pro": { + "format": "openai", + "flavor": "chat", + "multimodal": true, + "input_cost_per_mil_tokens": 2, + "output_cost_per_mil_tokens": 10, + "input_cache_read_cost_per_mil_tokens": 0.2, + "displayName": "Kimi K3.1 Pro", + "reasoning": true, + "max_input_tokens": 1000000, + "available_providers": [ + "databricks" + ] + }, + "moonshotai.kimi-k3.1-pro": { + "format": "converse", + "flavor": "chat", + "multimodal": true, + "input_cost_per_mil_tokens": 2, + "output_cost_per_mil_tokens": 10, + "input_cache_read_cost_per_mil_tokens": 0.2, + "displayName": "Kimi K3.1 Pro", + "reasoning": true, + "max_input_tokens": 1000000, + "available_providers": [ + "bedrock" + ] + }, + "us.moonshotai.kimi-k3.1-pro": { + "format": "converse", + "flavor": "chat", + "multimodal": true, + "input_cost_per_mil_tokens": 2, + "output_cost_per_mil_tokens": 10, + "input_cache_read_cost_per_mil_tokens": 0.2, + "displayName": "Kimi K3.1 Pro", + "reasoning": true, + "parent": "moonshotai.kimi-k3.1-pro", + "max_input_tokens": 1000000, + "available_providers": [ + "bedrock" + ] + }, + "global.moonshotai.kimi-k3.1-pro": { + "format": "converse", + "flavor": "chat", + "multimodal": true, + "input_cost_per_mil_tokens": 2, + "output_cost_per_mil_tokens": 10, + "input_cache_read_cost_per_mil_tokens": 0.2, + "displayName": "Kimi K3.1 Pro", + "reasoning": true, + "parent": "moonshotai.kimi-k3.1-pro", + "max_input_tokens": 1000000, + "available_providers": [ + "bedrock" + ] + }, "moonshotai/Kimi-K3": { "format": "openai", "flavor": "chat", @@ -24218,6 +24312,14 @@ "bedrock" ] }, + "wizardlm-2-8x22b": { + "format": "openai", + "flavor": "chat", + "input_cost_per_mil_tokens": 0.62, + "output_cost_per_mil_tokens": 0.62, + "displayName": "WizardLM-2 8x22B", + "available_providers": [] + }, "particle-ai/particle-1": { "format": "openai", "flavor": "chat", @@ -28047,13 +28149,5 @@ "available_providers": [ "openrouter" ] - }, - "wizardlm-2-8x22b": { - "format": "openai", - "flavor": "chat", - "input_cost_per_mil_tokens": 0.62, - "output_cost_per_mil_tokens": 0.62, - "displayName": "WizardLM-2 8x22B", - "available_providers": [] } }