Models
Every chat model behind the gateway, with what it can do, its context window, and what it costs per million tokens.
- Chat
- Streaming
- Image input
- Tool calling
- Reasoning
- Structured outputs
- Logprobs
- Prompt caching
- DeepSeek: DeepSeek V4 Flash262k contextChatMulti-turn conversations via /v1/chat/completions.ChatStreamingTokens arrive as they are generated with stream: true.StreamingTool callingNative tool calls, including parallel calls.Tool callingReasoningThinks before answering; the trace is billed as output.ReasoningStructured outputsJSON Schema responses; JSON mode on most models.Structured outputsLogprobsReturns token log-probabilities with logprobs: true.LogprobsPrompt cachingRepeated prompt prefixes bill at the cache-read rate.Prompt caching
DeepSeek V4 Flash is an open-weight reasoning model with a 262k context window, hosted by Reka. It is served over an OpenAI-compatible API with native tool calling, structured outputs, logprobs, and streaming.
- DeepSeek: DeepSeek V4 Pro1M contextChatMulti-turn conversations via /v1/chat/completions.ChatStreamingTokens arrive as they are generated with stream: true.StreamingTool callingNative tool calls, including parallel calls.Tool callingReasoningThinks before answering; the trace is billed as output.ReasoningStructured outputsJSON Schema responses; JSON mode on most models.Structured outputsLogprobsReturns token log-probabilities with logprobs: true.LogprobsPrompt cachingRepeated prompt prefixes bill at the cache-read rate.Prompt caching
DeepSeek V4 Pro is DeepSeek's flagship open-weight reasoning model with a 1M context window, hosted by Reka. It is served over an OpenAI-compatible API with native tool calling, structured outputs, logprobs, and streaming.
- GLM: GLM 5.3262k contextChatMulti-turn conversations via /v1/chat/completions.ChatStreamingTokens arrive as they are generated with stream: true.StreamingTool callingNative tool calls, including parallel calls.Tool callingReasoningThinks before answering; the trace is billed as output.ReasoningStructured outputsJSON Schema responses; JSON mode on most models.Structured outputsLogprobsReturns token log-probabilities with logprobs: true.LogprobsPrompt cachingRepeated prompt prefixes bill at the cache-read rate.Prompt caching
GLM 5.3 is an open-weight mixture-of-experts reasoning and coding model with a 262k context window, hosted by Reka. It is served over an OpenAI-compatible API with native tool calling, structured outputs, logprobs, and streaming.
- GLM: GLM 5.3 Flash262k contextChatMulti-turn conversations via /v1/chat/completions.ChatStreamingTokens arrive as they are generated with stream: true.StreamingImage inputAccepts images alongside text in the prompt.Image inputTool callingNative tool calls, including parallel calls.Tool callingReasoningThinks before answering; the trace is billed as output.ReasoningStructured outputsJSON Schema responses; JSON mode on most models.Structured outputsLogprobsReturns token log-probabilities with logprobs: true.LogprobsPrompt cachingRepeated prompt prefixes bill at the cache-read rate.Prompt caching
GLM 5.3 Flash is an open-weight 321B-parameter (18B active) reasoning model with a 262k context window, hosted by Reka. It is served over an OpenAI-compatible API with native tool calling, structured outputs, logprobs, and streaming.
- Qwen: Qwen3.8 27B262k contextChatMulti-turn conversations via /v1/chat/completions.ChatStreamingTokens arrive as they are generated with stream: true.StreamingImage inputAccepts images alongside text in the prompt.Image inputTool callingNative tool calls, including parallel calls.Tool callingReasoningThinks before answering; the trace is billed as output.ReasoningStructured outputsJSON Schema responses; JSON mode on most models.Structured outputsPrompt cachingRepeated prompt prefixes bill at the cache-read rate.Prompt caching
Qwen3.8 27B is an open-weight general-purpose model with a 262k context window and optional reasoning (set reasoning_effort to enable thinking), hosted by Reka. It is served over an OpenAI-compatible API with native tool calling, structured outputs, image input and streaming.
- Reka: Reka Edge 260316k contextChatMulti-turn conversations via /v1/chat/completions.ChatStreamingTokens arrive as they are generated with stream: true.StreamingImage inputAccepts images alongside text in the prompt.Image inputTool callingNative tool calls, including parallel calls.Tool callingStructured outputsJSON Schema responses; JSON mode on most models.Structured outputs
Reka Edge 2603 is Reka's model for physical AI, with a 16,384-token context window. It excels at visual understanding (videos, images), object detection, and tool use, and with its CONVNEXT vision encoder, it processes images with 3x fewer tokens, leading to significant gains in latency.
- Reka: Reka Flash 364k contextChatMulti-turn conversations via /v1/chat/completions.ChatStreamingTokens arrive as they are generated with stream: true.StreamingReasoningThinks before answering; the trace is billed as output.ReasoningStructured outputsJSON Schema responses; JSON mode on most models.Structured outputs
Reka Flash 3 is a 21B general-purpose reasoning LLM which excels at chat, coding, and instruction-following while maintaining a level of performance competitive with significantly larger models. Note: Reka Flash 3 is primarily an English model with limited multilingual understanding capabilities.
Note Hover an icon to see what the capability means, or use the legend at the top. The model id in each card is the exact string to pass as model; call GET /v1/models with your key for the live roster.