Skip to content

RAGFlow is an agent-native database: a Retrieval-Augmented Generation (RAG) engine that grew into context infrastructure.

feat: add Cheaper Inference as an OpenAI-compatible model provider (#19748)

aiapienthusiast46277ebe5c34
diff --git a/conf/llm_factories.json b/conf/llm_factories.json
index 172049ab65..3d68561295 100644
--- a/conf/llm_factories.json
+++ b/conf/llm_factories.json
@@ -2047,6 +2047,715 @@
"status": "1",
"llm": []
},
+ {
+ "name": "Cheaper Inference",
+ "logo": "",
+ "tags": "LLM,IMAGE2TEXT",
+ "status": "1",
+ "rank": "885",
+ "url": "https://api.cheaperinference.com/v1",
+ "llm": [
+ {
+ "llm_name": "aion-labs.aion-2-0",
+ "tags": "LLM,CHAT,128K",
+ "max_tokens": 128000,
+ "model_type": "chat",
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "claude-fable-5",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1000000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "claude-fable-5.1",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1000000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "claude-haiku-4.5",
+ "tags": "LLM,CHAT,200K,IMAGE2TEXT",
+ "max_tokens": 200000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools"
+ ]
+ },
+ {
+ "llm_name": "claude-opus-4-8-fast",
+ "tags": "LLM,CHAT,1M",
+ "max_tokens": 1000000,
+ "model_type": "chat",
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "claude-opus-4.5",
+ "tags": "LLM,CHAT,198K,IMAGE2TEXT",
+ "max_tokens": 198000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools"
+ ]
+ },
+ {
+ "llm_name": "claude-opus-4.6",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1000000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools"
+ ]
+ },
+ {
+ "llm_name": "claude-opus-4.7",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1000000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools"
+ ]
+ },
+ {
+ "llm_name": "claude-opus-4.8",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1000000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "claude-opus-5",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1000000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "claude-opus-5-fast",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1000000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "claude-sonnet-4.5",
+ "tags": "LLM,CHAT,198K,IMAGE2TEXT",
+ "max_tokens": 198000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools"
+ ]
+ },
+ {
+ "llm_name": "claude-sonnet-4.6",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1000000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools"
+ ]
+ },
+ {
+ "llm_name": "claude-sonnet-5",
+ "tags": "LLM,CHAT,1M",
+ "max_tokens": 1000000,
+ "model_type": "chat",
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "deepseek-v4-flash",
+ "tags": "LLM,CHAT,1M",
+ "max_tokens": 1000000,
+ "model_type": "chat",
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "deepseek-v4-flash-0731",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1048576,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "deepseek-v4-pro",
+ "tags": "LLM,CHAT,1M",
+ "max_tokens": 1000000,
+ "model_type": "chat",
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "deepseek-v4-pro-0813",
+ "tags": "LLM,CHAT,1M",
+ "max_tokens": 1048576,
+ "model_type": "chat",
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "deepseek-v4.1-flash",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1048576,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "gemini-2.5-flash",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1048576,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools"
+ ]
+ },
+ {
+ "llm_name": "gemini-3-5-flash",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1000000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "gemini-3-flash-preview",
+ "tags": "LLM,CHAT,256K,IMAGE2TEXT",
+ "max_tokens": 256000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "gemini-3.1-flash-lite",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1048576,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools"
+ ]
+ },
+ {
+ "llm_name": "gemini-3.1-pro",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1048576,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "gemini-3.1-pro-preview",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1000000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "gemini-3.7-flash",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1048576,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "glm-4.5",
+ "tags": "LLM,CHAT,128K",
+ "max_tokens": 128000,
+ "model_type": "chat",
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "glm-4.5-air",
+ "tags": "LLM,CHAT,128K",
+ "max_tokens": 128000,
+ "model_type": "chat",
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "glm-4.6",
+ "tags": "LLM,CHAT,198K",
+ "max_tokens": 198000,
+ "model_type": "chat",
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "glm-4.7",
+ "tags": "LLM,CHAT,128K",
+ "max_tokens": 128000,
+ "model_type": "chat",
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "glm-5",
+ "tags": "LLM,CHAT,198K",
+ "max_tokens": 198000,
+ "model_type": "chat",
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "glm-5.1",
+ "tags": "LLM,CHAT,200K",
+ "max_tokens": 200000,
+ "model_type": "chat",
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "glm-5.2",
+ "tags": "LLM,CHAT,1M",
+ "max_tokens": 1048576,
+ "model_type": "chat",
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "glm-5.3",
+ "tags": "LLM,CHAT,1M",
+ "max_tokens": 1000000,
+ "model_type": "chat",
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "glm-5.3-flash",
+ "tags": "LLM,CHAT,1M",
+ "max_tokens": 1048576,
+ "model_type": "chat",
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "gpt-5-mini",
+ "tags": "LLM,CHAT,400K",
+ "max_tokens": 400000,
+ "model_type": "chat",
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "gpt-5-nano",
+ "tags": "LLM,CHAT,400K",
+ "max_tokens": 400000,
+ "model_type": "chat",
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "gpt-5.2-codex",
+ "tags": "LLM,CHAT,256K,IMAGE2TEXT",
+ "max_tokens": 256000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "gpt-5.4",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1000000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "gpt-5.4-mini",
+ "tags": "LLM,CHAT,400K,IMAGE2TEXT",
+ "max_tokens": 400000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "gpt-5.4-nano",
+ "tags": "LLM,CHAT,400K",
+ "max_tokens": 400000,
+ "model_type": "chat",
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "gpt-5.5",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1000000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "gpt-5.6-luna",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1050000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "gpt-5.6-sol",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1050000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "gpt-5.6-terra",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1050000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "gpt-6-astra",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1050000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "grok-4.5",
+ "tags": "LLM,CHAT,500K,IMAGE2TEXT",
+ "max_tokens": 500000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "kimi-k3",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1000000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "minimax-m2.7",
+ "tags": "LLM,CHAT,198K",
+ "max_tokens": 198000,
+ "model_type": "chat",
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "muse-spark-1.2",
+ "tags": "LLM,CHAT,1M",
+ "max_tokens": 1048576,
+ "model_type": "chat",
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "qwen-3-8-27b",
+ "tags": "LLM,CHAT,262K,IMAGE2TEXT",
+ "max_tokens": 262144,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "qwen-3-8-max",
+ "tags": "LLM,CHAT,1M,IMAGE2TEXT",
+ "max_tokens": 1000000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "qwen3-5-35b-a3b",
+ "tags": "LLM,CHAT,256K,IMAGE2TEXT",
+ "max_tokens": 256000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "qwen3-6-35b-a3b",
+ "tags": "LLM,CHAT,262K",
+ "max_tokens": 262144,
+ "model_type": "chat",
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ },
+ {
+ "llm_name": "qwen3.6-27b",
+ "tags": "LLM,CHAT,256K,IMAGE2TEXT",
+ "max_tokens": 256000,
+ "model_type": [
+ "chat",
+ "vision"
+ ],
+ "is_tools": true,
+ "features": [
+ "is_tools",
+ "thinking"
+ ]
+ }
+ ]
+ },
{
"name": "DaoXE",
"logo": "",
diff --git a/conf/models/cheaperinference.json b/conf/models/cheaperinference.json
new file mode 100644
index 0000000000..67e686812b
--- /dev/null
+++ b/conf/models/cheaperinference.json
@@ -0,0 +1,841 @@
+{
+ "name": "Cheaper Inference",
+ "rank": 885,
+ "url": {
+ "default": "https://api.cheaperinference.com/v1"
+ },
+ "url_suffix": {
+ "chat": "chat/completions",
+ "models": "models"
+ },
+ "class": "cheaperinference",
+ "models": [
+ {
+ "name": "aion-labs.aion-2-0",
+ "context_length": 128000,
+ "max_output": 32768,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "claude-fable-5",
+ "context_length": 1000000,
+ "max_output": 128000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "claude-fable-5.1",
+ "context_length": 1000000,
+ "max_output": 128000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "claude-haiku-4.5",
+ "context_length": 200000,
+ "max_output": 64000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "claude-opus-4-8-fast",
+ "context_length": 1000000,
+ "max_output": 128000,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "claude-opus-4.5",
+ "context_length": 198000,
+ "max_output": 32768,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "claude-opus-4.6",
+ "context_length": 1000000,
+ "max_output": 128000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "claude-opus-4.7",
+ "context_length": 1000000,
+ "max_output": 128000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "claude-opus-4.8",
+ "context_length": 1000000,
+ "max_output": 128000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "claude-opus-5",
+ "context_length": 1000000,
+ "max_output": 128000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "claude-opus-5-fast",
+ "context_length": 1000000,
+ "max_output": 128000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "claude-sonnet-4.5",
+ "context_length": 198000,
+ "max_output": 64000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "claude-sonnet-4.6",
+ "context_length": 1000000,
+ "max_output": 64000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "claude-sonnet-5",
+ "context_length": 1000000,
+ "max_output": 64000,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "deepseek-v4-flash",
+ "context_length": 1000000,
+ "max_output": 128000,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "deepseek-v4-flash-0731",
+ "context_length": 1048576,
+ "max_output": 65536,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "deepseek-v4-pro",
+ "context_length": 1000000,
+ "max_output": 128000,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "deepseek-v4-pro-0813",
+ "context_length": 1048576,
+ "max_output": 384000,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "deepseek-v4.1-flash",
+ "context_length": 1048576,
+ "max_output": 384000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "gemini-2.5-flash",
+ "context_length": 1048576,
+ "max_output": 65535,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "gemini-3-5-flash",
+ "context_length": 1000000,
+ "max_output": 65536,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "gemini-3-flash-preview",
+ "context_length": 256000,
+ "max_output": 65536,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "gemini-3.1-flash-lite",
+ "context_length": 1048576,
+ "max_output": 65536,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "gemini-3.1-pro",
+ "context_length": 1048576,
+ "max_output": 65536,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "gemini-3.1-pro-preview",
+ "context_length": 1000000,
+ "max_output": 32768,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "gemini-3.7-flash",
+ "context_length": 1048576,
+ "max_output": 65536,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "glm-4.5",
+ "context_length": 128000,
+ "max_output": 32768,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "glm-4.5-air",
+ "context_length": 128000,
+ "max_output": 32768,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "glm-4.6",
+ "context_length": 198000,
+ "max_output": 16384,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "glm-4.7",
+ "context_length": 128000,
+ "max_output": 32768,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "glm-5",
+ "context_length": 198000,
+ "max_output": 32768,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "glm-5.1",
+ "context_length": 200000,
+ "max_output": 24000,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "glm-5.2",
+ "context_length": 1048576,
+ "max_output": 131072,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "glm-5.3",
+ "context_length": 1000000,
+ "max_output": 131072,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "glm-5.3-flash",
+ "context_length": 1048576,
+ "max_output": 131072,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "gpt-5-mini",
+ "context_length": 400000,
+ "max_output": 128000,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "gpt-5-nano",
+ "context_length": 400000,
+ "max_output": 128000,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "gpt-5.2-codex",
+ "context_length": 256000,
+ "max_output": 65536,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "gpt-5.4",
+ "context_length": 1000000,
+ "max_output": 131072,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "gpt-5.4-mini",
+ "context_length": 400000,
+ "max_output": 128000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "gpt-5.4-nano",
+ "context_length": 400000,
+ "max_output": 128000,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "gpt-5.5",
+ "context_length": 1000000,
+ "max_output": 128000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "gpt-5.6-luna",
+ "context_length": 1050000,
+ "max_output": 128000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "gpt-5.6-sol",
+ "context_length": 1050000,
+ "max_output": 128000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "gpt-5.6-terra",
+ "context_length": 1050000,
+ "max_output": 128000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "gpt-6-astra",
+ "context_length": 1050000,
+ "max_output": 128000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "grok-4.5",
+ "context_length": 500000,
+ "max_output": 128000,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "kimi-k3",
+ "context_length": 1000000,
+ "max_output": 131072,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "minimax-m2.7",
+ "context_length": 198000,
+ "max_output": 32768,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "muse-spark-1.2",
+ "context_length": 1048576,
+ "max_output": 943718,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "qwen-3-8-27b",
+ "context_length": 262144,
+ "max_output": 65536,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "qwen-3-8-max",
+ "context_length": 1000000,
+ "max_output": 131072,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "qwen3-5-35b-a3b",
+ "context_length": 256000,
+ "max_output": 65536,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "qwen3-6-35b-a3b",
+ "context_length": 262144,
+ "max_output": 262144,
+ "model_types": [
+ "chat"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ },
+ {
+ "name": "qwen3.6-27b",
+ "context_length": 256000,
+ "max_output": 32768,
+ "model_types": [
+ "chat",
+ "vision"
+ ],
+ "thinking": {
+ "default_value": true,
+ "clear_thinking": true
+ },
+ "tools": {
+ "support": true
+ }
+ }
+ ]
+}
diff --git a/docs/guides/models/supported_models.mdx b/docs/guides/models/supported_models.mdx
index ee2cb9903f..544cbf959a 100644
--- a/docs/guides/models/supported_models.mdx
+++ b/docs/guides/models/supported_models.mdx
@@ -31,6 +31,7 @@ A complete list of model providers supported by RAGFlow, which will continue to
| BaiChuan | `https://www.baichuan-ai.com` |
| BaiduYiyan | `https://yiyan.baidu.com` |
| Bedrock | `https://aws.amazon.com/bedrock/` |
+| Cheaper Inference | `https://cheaperinference.com` |
| Cohere | `https://cohere.com` |
| CometAPI | `https://cometapi.com` |
| DaoXE | `https://daoxe.com` |
diff --git a/internal/entity/models/cheaperinference.go b/internal/entity/models/cheaperinference.go
new file mode 100644
index 0000000000..01e60a162e
--- /dev/null
+++ b/internal/entity/models/cheaperinference.go
@@ -0,0 +1,45 @@
+//
+// Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+
+package models
+
+// CheaperInferenceModel implements Cheaper Inference's OpenAI-compatible chat
+// and model-list APIs.
+//
+// Responses carry an extra "provider" member naming the upstream that served
+// the request. The shared response decoder ignores members it does not know, so
+// the embedded OpenAIAPICompatibleModel covers every call and nothing is
+// overridden beyond the driver's own identity.
+type CheaperInferenceModel struct {
+ *OpenAIAPICompatibleModel
+}
+
+// NewCheaperInferenceModel creates a Cheaper Inference model driver.
+func NewCheaperInferenceModel(baseURL map[string]string, urlSuffix URLSuffix) *CheaperInferenceModel {
+ return &CheaperInferenceModel{
+ OpenAIAPICompatibleModel: NewOpenAIAPICompatibleModel(baseURL, urlSuffix),
+ }
+}
+
+// Name returns the model driver name.
+func (m *CheaperInferenceModel) Name() string {
+ return "Cheaper Inference"
+}
+
+// NewInstance creates a new Cheaper Inference driver bound to the given base URL.
+func (m *CheaperInferenceModel) NewInstance(baseURL map[string]string) ModelDriver {
+ return NewCheaperInferenceModel(baseURL, m.baseModel.URLSuffix)
+}
diff --git a/internal/entity/models/cheaperinference_test.go b/internal/entity/models/cheaperinference_test.go
new file mode 100644
index 0000000000..886f36d33a
--- /dev/null
+++ b/internal/entity/models/cheaperinference_test.go
@@ -0,0 +1,98 @@
+package models
+
+import (
+ "encoding/json"
+ "net/http"
+ "net/http/httptest"
+ "ragflow/internal/common"
+ "testing"
+)
+
+func newCheaperInferenceForTest(baseURL string) *CheaperInferenceModel {
+ return NewCheaperInferenceModel(
+ map[string]string{"default": baseURL},
+ URLSuffix{Chat: "chat/completions", Models: "models"},
+ )
+}
+
+func TestCheaperInferenceName(t *testing.T) {
+ if got := newCheaperInferenceForTest("http://unused").Name(); got != "Cheaper Inference" {
+ t.Errorf("Name()=%q", got)
+ }
+}
+
+func TestCheaperInferenceFactory(t *testing.T) {
+ driver, err := NewModelFactory().CreateModelDriver("Cheaper Inference", map[string]string{"default": "http://unused"}, URLSuffix{})
+ if err != nil {
+ t.Fatalf("CreateModelDriver: %v", err)
+ }
+ if _, ok := driver.(*CheaperInferenceModel); !ok {
+ t.Fatalf("driver type=%T, want *CheaperInferenceModel", driver)
+ }
+ if _, ok := driver.NewInstance(map[string]string{"default": "http://other"}).(*CheaperInferenceModel); !ok {
+ t.Fatal("NewInstance did not return *CheaperInferenceModel")
+ }
+}
+
+// Cheaper Inference is a hosted gateway, so unlike a local server it must
+// receive the tenant's key as a bearer token on every call.
+func TestCheaperInferenceChatSendsAPIKey(t *testing.T) {
+ withSSRFBypass(t)
+ srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+ if r.URL.Path != "/chat/completions" {
+ t.Errorf("path=%s", r.URL.Path)
+ }
+ if got, want := r.Header.Get("Authorization"), "Bearer "+testAPIKey; got != want {
+ t.Errorf("Authorization=%q, want %q", got, want)
+ }
+ _ = json.NewEncoder(w).Encode(map[string]any{
+ "id": "chat-cheaperinference",
+ "choices": []map[string]any{{"message": map[string]any{"content": "pong"}}},
+ "usage": map[string]any{"prompt_tokens": 3, "completion_tokens": 5, "total_tokens": 8},
+ })
+ }))
+ defer srv.Close()
+
+ usage := &common.ModelUsage{}
+ resp, err := newCheaperInferenceForTest(srv.URL).ChatWithMessages(
+ t.Context(), "claude-opus-5", []Message{{Role: "user", Content: "ping"}},
+ &APIConfig{ApiKey: &testAPIKey}, nil, usage,
+ )
+ if err != nil {
+ t.Fatalf("ChatWithMessages: %v", err)
+ }
+ if *resp.Answer != "pong" {
+ t.Errorf("Answer=%q", *resp.Answer)
+ }
+ assertModelUsage(t, usage, 3, 5, 8)
+}
+
+// Responses carry extra members the decoder does not know. They must not disturb
+// answer or usage extraction.
+func TestCheaperInferenceChatAcceptsUnknownResponseMembers(t *testing.T) {
+ withSSRFBypass(t)
+ srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
+ _ = json.NewEncoder(w).Encode(map[string]any{
+ "id": "chat-cheaperinference-extra-members",
+ "model": "claude-opus-5",
+ "provider": "upstream-provider-name",
+ "service_tier": "default",
+ "choices": []map[string]any{{"message": map[string]any{"content": "pong"}}},
+ "usage": map[string]any{"prompt_tokens": 3, "completion_tokens": 5, "total_tokens": 8},
+ })
+ }))
+ defer srv.Close()
+
+ usage := &common.ModelUsage{}
+ resp, err := newCheaperInferenceForTest(srv.URL).ChatWithMessages(
+ t.Context(), "claude-opus-5", []Message{{Role: "user", Content: "ping"}},
+ &APIConfig{ApiKey: &testAPIKey}, nil, usage,
+ )
+ if err != nil {
+ t.Fatalf("ChatWithMessages: %v", err)
+ }
+ if *resp.Answer != "pong" {
+ t.Errorf("Answer=%q", *resp.Answer)
+ }
+ assertModelUsage(t, usage, 3, 5, 8)
+}
diff --git a/internal/entity/models/factory.go b/internal/entity/models/factory.go
index 6e8a8d9474..3729c50003 100644
--- a/internal/entity/models/factory.go
+++ b/internal/entity/models/factory.go
@@ -83,6 +83,8 @@ func (f *ModelFactory) CreateModelDriver(providerName string, baseURL map[string
return NewHuggingFaceModel(baseURL, urlSuffix), nil
case "baiduyiyan":
return NewBaiduModel(baseURL, urlSuffix), nil
+ case "cheaper inference":
+ return NewCheaperInferenceModel(baseURL, urlSuffix), nil
case "cohere":
return NewCoHereModel(baseURL, urlSuffix), nil
case "cometapi":
diff --git a/rag/llm/chat_model.py b/rag/llm/chat_model.py
index 5dff2b1f8f..684ac7fcde 100644
--- a/rag/llm/chat_model.py
+++ b/rag/llm/chat_model.py
@@ -3066,6 +3066,51 @@ class RAGconChat(Base):
super().__init__(key, model_name, base_url, **kwargs)
+class CheaperInferenceChat(Base):
+ """Cheaper Inference OpenAI-compatible chat adapter.
+
+ Cheaper Inference is a hosted OpenAI-compatible gateway, so the standard
+ OpenAI client path covers every call. A reply carries an extra ``provider``
+ member naming the upstream that served it; the OpenAI client keeps it out of
+ the way of answer and usage extraction, so it needs no handling here. The
+ base URL stays tenant-configurable because the gateway is also reachable
+ through per-account domains.
+ """
+
+ _FACTORY_NAME = "Cheaper Inference"
+
+ def __init__(self, key, model_name, base_url, **kwargs):
+ if not base_url:
+ raise ValueError("url cannot be None")
+ model_name = model_name.split("___")[0]
+ super().__init__(key, model_name, base_url, **kwargs)
+
+ def _clean_conf(self, gen_conf):
+ """Map the thinking selection onto the field the gateway reads.
+
+ The gateway reads the OpenAI ``reasoning_effort`` field and ignores
+ ``thinking`` and ``enable_thinking``. Measured on claude-opus-5 with
+ one prompt: no field gives 78-80 reasoning tokens, ``high`` 44-63,
+ ``medium`` 31-41, ``low`` 24-25 and ``none`` 0, while
+ ``thinking={"type": "disabled"}`` keeps the default 56. ``none`` is
+ accepted by every model family the gateway serves. ``Base._clean_conf``
+ drops all three keys, so the translation happens here.
+ """
+ thinking = gen_conf.get("thinking")
+ if isinstance(thinking, dict):
+ thinking = thinking.get("type")
+ enable_thinking = gen_conf.get("enable_thinking")
+ reasoning_effort = gen_conf.get("reasoning_effort")
+
+ gen_conf = super()._clean_conf(gen_conf)
+
+ if thinking == "disabled" or enable_thinking is False:
+ reasoning_effort = "none"
+ if reasoning_effort:
+ gen_conf["reasoning_effort"] = reasoning_effort
+ return gen_conf
+
+
class DaoXEChat(Base):
"""DaoXE OpenAI-compatible chat adapter.
diff --git a/rag/llm/cv_model.py b/rag/llm/cv_model.py
index cf968e6393..71180850ea 100644
--- a/rag/llm/cv_model.py
+++ b/rag/llm/cv_model.py
@@ -1456,3 +1456,26 @@ class NewAPICv(GptV4):
self.model_name = model_name.split("___")[0]
self.lang = lang
Base.__init__(self, **kwargs)
+
+
+class CheaperInferenceCV(GptV4):
+ """Cheaper Inference vision adapter.
+
+ The gateway takes image input on the same OpenAI-compatible
+ ``chat/completions`` route it uses for text, so the standard OpenAI client
+ path covers ``describe`` and ``describe_with_prompt`` with nothing
+ overridden. The base URL stays tenant-configurable because the gateway is
+ also reachable through per-account domains.
+ """
+
+ _FACTORY_NAME = "Cheaper Inference"
+
+ def __init__(self, key, model_name, lang="Chinese", base_url="", **kwargs):
+ if not base_url:
+ raise ValueError("url cannot be None")
+ self.base_url = ensure_v1(base_url)
+ self.client = OpenAI(api_key=key, base_url=self.base_url)
+ self.async_client = AsyncOpenAI(api_key=key, base_url=self.base_url)
+ self.model_name = model_name.split("___")[0]
+ self.lang = lang
+ Base.__init__(self, **kwargs)
diff --git a/rag/llm/model_meta.py b/rag/llm/model_meta.py
index 00e7109007..f11b61b26f 100644
--- a/rag/llm/model_meta.py
+++ b/rag/llm/model_meta.py
@@ -1187,6 +1187,77 @@ class ApiRoute(OpenAIAPICompatible):
return f"{self._BASE_URL}/models"
+class CheaperInference(OpenAIAPICompatible):
+ """Cheaper Inference catalog lister.
+
+ ``conf/models/cheaperinference.json`` pins the catalog the gateway
+ documents, which is what the model pickers show. This lister covers the
+ on-demand refresh: it reads the gateway's own ``/v1/models`` endpoint so
+ routes added after this file shipped are still discoverable against the
+ tenant's own key.
+
+ The listing is richer than the OpenAI shape the parent assumes: each entry
+ names the endpoint that serves it, its modality and its own capability
+ flags. The parent infers model types from the model id, which would file
+ the gateway's image-generation and video routes as chat models and would
+ miss image input on every id that carries no ``vl``/``vision`` hint, so the
+ entry's own fields are read instead. The endpoint publishes no tool-calling
+ flag, so ``is_tools`` follows the pinned catalog and is set for every chat
+ model.
+ """
+
+ _FACTORY_NAME = "Cheaper Inference"
+
+ _CHAT_ENDPOINT = "/v1/chat/completions"
+ _CHAT_MODALITY = "text"
+
+ def _format_model_list(self, raw_model_list):
+ models = raw_model_list.get("data") if isinstance(raw_model_list, dict) else raw_model_list
+ if not isinstance(models, list):
+ return []
+
+ model_list = []
+ for model in models:
+ if not isinstance(model, dict):
+ continue
+
+ model_name = model.get("id") or model.get("name")
+ if not model_name:
+ continue
+
+ endpoint = model.get("endpoint")
+ if endpoint and endpoint != self._CHAT_ENDPOINT:
+ continue
+ modality = model.get("type")
+ if modality and modality != self._CHAT_MODALITY:
+ continue
+
+ capabilities = model.get("capabilities")
+ if not isinstance(capabilities, dict):
+ capabilities = {}
+ model_types = [LLMType.CHAT.value]
+ if capabilities.get("vision"):
+ model_types.append(LLMType.VISION.value)
+ features = ["is_tools"]
+ if capabilities.get("reasoning"):
+ features.append("thinking")
+
+ context_length = model.get("context_length")
+ if not isinstance(context_length, int) or isinstance(context_length, bool) or context_length <= 0:
+ context_length = 8192
+
+ model_list.append(
+ {
+ "name": model_name,
+ "model_types": model_types,
+ "features": features,
+ "max_tokens": context_length,
+ }
+ )
+
+ return model_list
+
+
class DaoXE(OpenAIAPICompatible):
"""DaoXE catalog lister.
diff --git a/test/unit_test/rag/llm/test_cheaperinference_reasoning_controls.py b/test/unit_test/rag/llm/test_cheaperinference_reasoning_controls.py
new file mode 100644
index 0000000000..e9cbd48b67
--- /dev/null
+++ b/test/unit_test/rag/llm/test_cheaperinference_reasoning_controls.py
@@ -0,0 +1,84 @@
+#
+# Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+"""
+The Cheaper Inference gateway reads ``reasoning_effort`` and ignores
+``thinking`` / ``enable_thinking``, so ``CheaperInferenceChat._clean_conf``
+translates the RAGFlow thinking selection into ``reasoning_effort``.
+``Base._clean_conf`` drops all three keys, which would silently discard an
+explicit selection.
+"""
+
+import pytest
+
+from rag.llm.chat_model import CheaperInferenceChat
+
+pytestmark = pytest.mark.p2
+
+BASE_URL = "https://api.cheaperinference.com/v1"
+
+
+def _chat():
+ return CheaperInferenceChat("ci-test", "claude-opus-5", base_url=BASE_URL)
+
+
+def test_thinking_disabled_turns_reasoning_off():
+ cleaned = _chat()._clean_conf({"temperature": 0.2, "thinking": "disabled"})
+
+ assert cleaned["reasoning_effort"] == "none"
+ assert cleaned["temperature"] == 0.2
+ assert "thinking" not in cleaned
+
+
+def test_thinking_disabled_in_dict_form_turns_reasoning_off():
+ cleaned = _chat()._clean_conf({"thinking": {"type": "disabled"}})
+
+ assert cleaned["reasoning_effort"] == "none"
+
+
+def test_enable_thinking_false_turns_reasoning_off():
+ cleaned = _chat()._clean_conf({"enable_thinking": False})
+
+ assert cleaned["reasoning_effort"] == "none"
+ assert "enable_thinking" not in cleaned
+
+
+def test_thinking_enabled_keeps_the_model_default():
+ # The gateway has no "force on" field; a reasoning model reasons by
+ # default, so nothing is added.
+ cleaned = _chat()._clean_conf({"thinking": "enabled"})
+
+ assert "reasoning_effort" not in cleaned
+ assert "thinking" not in cleaned
+
+
+def test_explicit_reasoning_effort_survives():
+ cleaned = _chat()._clean_conf({"reasoning_effort": "low", "thinking": "enabled"})
+
+ assert cleaned["reasoning_effort"] == "low"
+
+
+def test_thinking_disabled_overrides_an_explicit_effort():
+ cleaned = _chat()._clean_conf({"reasoning_effort": "high", "thinking": "disabled"})
+
+ assert cleaned["reasoning_effort"] == "none"
+
+
+def test_no_selection_adds_nothing():
+ cleaned = _chat()._clean_conf({"temperature": 0.7, "max_tokens": 100})
+
+ assert "reasoning_effort" not in cleaned
+ assert "max_tokens" not in cleaned
diff --git a/web/src/assets/svg/llm/cheaperinference.svg b/web/src/assets/svg/llm/cheaperinference.svg
new file mode 100644
index 0000000000..4f47d046bc
--- /dev/null
+++ b/web/src/assets/svg/llm/cheaperinference.svg
@@ -0,0 +1,5 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="48" height="48" viewBox="0 0 64 64">
+ <rect x=".75" y=".75" width="62.5" height="62.5" rx="13.5" fill="#fff" stroke="#040d10" stroke-opacity=".09"/>
+ <path fill="#040d10" d="M38.25 15.75H27.4C18.07 15.75 10.5 23.03 10.5 32c0 8.97 7.52 16.25 16.85 16.25l10.9-10.7h-9.7a5.57 5.57 0 0 1-5.6-5.55c0-3.1 2.5-5.6 5.6-5.6h9.7V15.75Z"/>
+ <path fill="#040d10" d="m41.4 25.55 7.74-8.1c1.48-1.55 4.11-.5 4.11 1.64v26.59a2.57 2.57 0 0 1-2.57 2.57H41.4v-22.7Z"/>
+</svg>
diff --git a/web/src/components/svg-icon.tsx b/web/src/components/svg-icon.tsx
index 7763c886ec..6403a63bb6 100644
--- a/web/src/components/svg-icon.tsx
+++ b/web/src/components/svg-icon.tsx
@@ -116,6 +116,7 @@ const svgIcons = [
LLMFactory.Synthorai,
LLMFactory.ApiRoute,
LLMFactory.MWS,
+ LLMFactory.CheaperInference,
];
export const LlmIcon = ({
diff --git a/web/src/constants/llm.ts b/web/src/constants/llm.ts
index 367797cafe..f30a4ad894 100644
--- a/web/src/constants/llm.ts
+++ b/web/src/constants/llm.ts
@@ -109,6 +109,7 @@ export enum LLMFactory {
GreenPT = 'GreenPT',
Synthorai = 'Synthorai',
ApiRoute = 'API-Route',
+ CheaperInference = 'Cheaper Inference',
}
// Please lowercase the file name
@@ -201,6 +202,7 @@ export const IconMap = {
[LLMFactory.Synthorai]: 'synthorai',
[LLMFactory.ApiRoute]: 'apiroute',
[LLMFactory.MWS]: 'mws',
+ [LLMFactory.CheaperInference]: 'cheaperinference',
};
export const ModelTypeToField: Record<string, string> = {
@@ -228,6 +230,7 @@ export const APIMapUrl = {
[LLMFactory.Synthorai]: 'https://synthorai.io',
[LLMFactory.ApiRoute]: 'https://www.api-route.com',
[LLMFactory.DaoXE]: 'https://daoxe.com',
+ [LLMFactory.CheaperInference]: 'https://cheaperinference.com',
[LLMFactory.MWS]:
'https://mws.ru/docs/cloud-platform/gpt/general/inference-text.html',
[LLMFactory.Anthropic]: 'https://console.anthropic.com/settings/keys',