SiliconFlow (BYOK)BYOK

DeepSeek-R1 is DeepSeek's open-weight reasoning model, a 671B mixture-of-experts with 37B active parameters and a 131,072-token context. It is trained with large-scale reinforcement learning and released under the MIT license. It targets math, code and reasoning.

DeepSeek V3.1 Terminus is an open-weight update to DeepSeek V3.1. It keeps the original capabilities while addressing user reports on language consistency and agent capabilities. It supports reasoning, streaming and function calling.

DeepSeek V3.1 is an open-weight hybrid reasoning model from DeepSeek with a 131,072-token context window, strong coding and agentic tool-use performance. Served free through SambaNova Cloud's fast RDU inference and SiliconFlow.

DeepSeek V3.2 Exp is an experimental DeepSeek model built on V3.1-Terminus that adds DeepSeek Sparse Attention to improve long-context training and inference efficiency. It matches V3.1-Terminus on public benchmarks, and is released with open weights under the MIT license.

DeepSeek-V3 is DeepSeek's 671B mixture-of-experts general-purpose model. It handles multilingual, coding and reasoning tasks. It is a text model with streaming and function calling.

DeepSeek-V4-Flash is DeepSeek's open-weight fast mixture-of-experts model for high-throughput chat and coding. It accepts text and image input and returns text. It supports reasoning, tool calling and agentic use.

DeepSeek V4 Pro is DeepSeek's open-weight high-capability reasoning model. It is a text model for chat and coding with streaming and function calling.

DeepSeek V4.1 Flash 1017 is the October 17, 2026 snapshot of DeepSeek V4.1 Flash. It is an open-weight sparse mixture-of-experts model with native vision input, text output and a 1M-token context window.

DeepSeek V4.1 Flash is the open-weight, cost-efficient sparse mixture-of-experts tier of the V4.1 family. It has 552B total parameters with 8B active on input and 16B on output, 1M context, and native vision in with text out. This is the original September 2026 checkpoint.

Gemma 4 26B A4B is an open-weight model from Google's Gemma 4 family, built from Gemini 3 research to maximize intelligence per parameter. It takes text and image input and generates text. It supports reasoning, tool calling, and coding.

Gemma 4 31B is Google's open-weight dense open reasoning model for complex reasoning, agentic workflows, coding, and multimodal understanding. It takes text and image input and generates text. It supports tool calling and a thinking mode.

Kev 4B is Jared Palmer's small open-weight decision model, a LoRA adapter and pointer head on Qwen3.5-4B-Base. It returns a calibrated probability per typed question in one forward pass. It does not generate text.

LongCat-2.0 is Meituan's open-weight sparse mixture-of-experts language model with 1.6T total and 48B active parameters. It targets coding, repository-level changes, and agentic workflows. It supports streaming, function calling, and structured output.

MiniMax M3 is MiniMax's multimodal language model for coding and agentic work. It accepts text, image and video input and returns text. It supports reasoning, function calling, structured output and multilingual use.

MiniMax M3.1 Flash is MiniMax's open-weight multimodal model for complex agentic tasks, coding and cybersecurity. It accepts text, image and video input and has a 1M-token context window. It always thinks, with adjustable effort, and its weights are not yet released.

Kimi K2.5 is Moonshot AI's open-weight mixture-of-experts model with native vision, tool calling and structured outputs. It has 1T total and 32B active parameters, a 256K context window and a Modified MIT license.

Kimi K2.6 is Moonshot AI's open-weight large mixture-of-experts model with agentic tool use, vision input and long-context memory. It supports reasoning and streaming, and takes text and image input.

Kimi K2.7 Code is Moonshot AI's open-weight coding-focused model for end-to-end programming tasks over long contexts. It is a native multimodal mixture-of-experts model with 32B active of roughly 1T total parameters and always-on thinking.

Kimi K3.1 Pro is the first model in Moonshot AI's open-weight K3.1 series, a post-trained version of Kimi K3. It has a 1M-token context window and native visual understanding, and improves on K3 in coding, 3D and CAD modeling and financial research. Thinking is always on.

Kimi K3 is Moonshot AI's open-weight multimodal reasoning model with native visual understanding. It accepts text, image and video input for long-horizon software engineering, knowledge work and CAD. Thinking is always on.

GPT-OSS-120b is OpenAI's open-weight mixture-of-experts reasoning model under the Apache-2.0 license. It has 116.8B total and 5.1B active parameters and a 131,072-token context window. It supports function calling and runs on one 80 GB GPU with MXFP4 weights.

GPT-OSS 20B is OpenAI's open-weight reasoning model for agentic tasks and low-latency edge use, under Apache-2.0. It is a mixture-of-experts model with 20.9B total and 3.6B active parameters and a 131,072-token context. It runs within 16 GB of memory with MXFP4 weights.

Qwen2.5-72B-Instruct is Alibaba Qwen's open-weight 72.7B-parameter general chat model. It is stronger at coding and maths than Qwen2, supports 29+ languages and handles function calling.

Qwen3-14B is an open-weight 14.8B-parameter dense model from Alibaba Qwen that balances quality and latency. It supports reasoning, function calling, 100+ languages and dialects, instruction following and translation.

Qwen3 30B A3B Instruct 2507 is Qwen's open-weight 30.5B-parameter mixture-of-experts language model with 3.3B active parameters per inference. It has a 262,144-token context window and supports function calling and streaming.

Qwen3-8B is Alibaba Qwen's open-weight small dense model with 8B parameters. It is fast and efficient, supports reasoning, and suits lightweight chat and tool use.

Qwen3 Embedding 4B is the open-weight 4B size in the Qwen3 embedding and reranking family. It embeds text over a 32,768-token window at 2,560 dimensions, and supports 100+ languages and Matryoshka truncation from 32 to 2,560 dimensions.

Qwen3 Embedding 8B is Qwen's open-weight 8B text embedding model with a 32,768-token context window. It takes text in and returns embedding vectors.

Qwen3 VL 30B A3B Instruct is Qwen's open-weight multimodal model that combines text generation with visual understanding of images and videos. It targets instruction following and supports function calling.

Qwen3.5-122B-A10B is Qwen's open-weight native vision-language model with a hybrid architecture. It combines linear attention with a sparse mixture-of-experts, takes text, image and video input.

Qwen3.5-27B is Alibaba's largest open-weight dense Qwen3.5 model. It has a thinking mode, tool calling, multi-token prediction and 201 languages, with text, image and video input.

Qwen3.5-35B-A3B is Qwen's open-weight native vision-language model with a hybrid of linear attention and a sparse mixture-of-experts. It takes text, image and video input, has a 262,144-token context window and supports reasoning.

Qwen3.5-9B is an open-weight model from Alibaba's Qwen3.5 series with a hybrid Gated Delta Networks and sparse MoE architecture. It has a thinking mode, tool calling and 201 languages.

Qwen3.8 2.4T A95B is Qwen's sparse mixture-of-experts model, the open-weight variant of Qwen3.8 Max. It has 95 billion active of 2.4 trillion total parameters.

Qwen3.8-27B is Alibaba's open-weight dense Qwen3.8 model for agentic coding and complex reasoning, with a native thinking mode. It targets multi-step agent work and large coding tasks.

Step 3.5 Flash is StepFun's open-weight sparse Mixture-of-Experts model with 196.81B total parameters and about 11B active per token. It uses a 45-layer transformer with 288 routed experts, Top-8 selection, and 256K context for reasoning and agentic tasks.

Hy4 Preview is Tencent Hunyuan's open-weight 770B mixture-of-experts model with 49B active parameters, built for agents and coding. It targets office automation and complex tool use.

GLM 4.5 Air is the open-weight, lightweight variant of Z.ai's GLM-4.5 flagship family, built for agent-centric applications. It supports reasoning, function calling, and streaming, with a 131,072-token context window.

GLM-5.1 is Z-AI's open-weight flagship foundation model for long-horizon autonomous tasks. It has a 128K output window and scores 58.4 on SWE-Bench Pro. It supports thinking mode, function calling, and MCP integration.

GLM-5.2 is Z-AI's open-weight flagship reasoning model with a 128K output window. It supports controllable thinking effort, function calling, and MCP integration. It targets long-horizon autonomous coding and agentic tasks.

GLM-5.3 is Z.AI's open-weight reasoning model for coding and agentic workflows. It targets complex software engineering, long-running agents, and vulnerability analysis. It uses the same base model as GLM-5.2 with scaled post-training for stronger coding and token efficiency.

GLM-5 is Z-AI's open-weight general-purpose foundation model for reasoning, coding, and agentic workloads. It has a 128K context window and 32K output. It supports thinking mode, function calling, and structured outputs.

GLM 5V Turbo is Z.ai's first native multimodal agent foundation model, built for vision-based coding and agent-driven tasks. It takes image, text, and video input and returns text, with a 202,752-token context window.