Open-Source LLM APIs
Hosted APIs for open-weight language models: DeepSeek, Qwen, Kimi, GLM, MiniMax, MiMo, Hunyuan, gpt-oss and Llama, without running your own GPUs.
All Open-Source LLM APIs
Preview the output style, compare capabilities, then open any model for full specs, pricing, and a live Playground.
DeepSeek V4 Pro
DeepSeek V4 Pro is the large tier of DeepSeek's V4 family, a 1.6T-parameter mixture-of-experts model with 49B active parameters, introduced as an open-weight preview in April 2026 with a 1M-token cont...
DeepSeek V3.2
DeepSeek-V3.2 is DeepSeek's open-weight, reasoning-first model released in December 2025 as the official successor to V3.2-Exp. It introduces DeepSeek Sparse Attention to reduce long-context training ...
Qwen3.8 2.4T A95B
Qwen3.8 2.4T A95B is the open-weight release of Alibaba's Qwen3.8 Max-class model, a sparse Mixture-of-Experts with 2.4 trillion total and about 95 billion active parameters, published in August 2026 ...
Qwen3.8 27B
Qwen3.8 27B is the compact dense member of Alibaba's Qwen3.8 family, an open-weight native vision-language model that understands images and video, with thinking mode on by default. Compared with Qwen...
Qwen3.6 27B
Qwen3.6 27B is an open-weight dense 27-billion-parameter vision-language model in Alibaba's Qwen3.6 family, released in April 2026 with thinking and non-thinking modes. Qwen positions it for agentic c...
Qwen3.5-35B-A3B
Qwen3.5 35B A3B is an open-weight native vision-language model from Alibaba's Qwen3.5 series, released in February 2026, pairing linear attention with a sparse Mixture-of-Experts design that activates...
Kimi K3.1 Pro
Kimi K3.1 Pro is Moonshot AI's flagship open-weight multimodal reasoning model, a post-trained successor to Kimi K3 with a 1M-token context window. It accepts text, image and video input and is built ...
Kimi K3
Kimi K3 is Moonshot AI's flagship model, a 2.8-trillion-parameter open-weight mixture-of-experts model with native visual understanding and a 1M-token context. It is built on Kimi Delta Attention, a h...
GLM 5.2
GLM-5.2 is Z.ai's open-weight flagship text model released in June 2026 for long-horizon tasks, and the first GLM model to deliver that capability on a usable 1M-token context. Z.ai positions it for p...
MiniMax M3.1 Flash
MiniMax-M3.1-Flash is the fast, cost-efficient tier of MiniMax's M3.1 generation, which MiniMax describes as an ultra-efficient open-weight model for everyday work. It is a multimodal coding model tha...
MiniMax M3
MiniMax-M3 is MiniMax's flagship multimodal foundation model, which MiniMax describes as open-weight. It is built on MiniMax Sparse Attention, which selects key-value blocks to cut long-context comput...
MiMo-V2.6-Pro
MiMo-V2.6-Pro is the flagship of Xiaomi's open-source MiMo-V2.6 series, a trillion-parameter, natively omni-modal reasoning model released in September 2026 that takes text, image, video and audio inp...
MiMo-V2.6-Flash
MiMo-V2.6-Flash is the efficiency tier of Xiaomi's open-source MiMo-V2.6 series, released in September 2026 as a natively omni-modal reasoning model that takes text, image, video and audio input and r...
Hy4 preview
Hy4 preview is Tencent Hunyuan's open-source preview of its Hy4 generation, a mixture-of-experts language model with 770B total and 49B active parameters and a context window exceeding 1M tokens. Tenc...
gpt-oss-120b
gpt-oss-120b is OpenAI's largest open-weight reasoning model, released in August 2025 under the Apache 2.0 license: a 117B-parameter mixture-of-experts model with 5.1B active parameters that fits on a...
Llama 3.3 70B Instruct
Llama 3.3 70B Instruct is Meta's instruction-tuned, text-only multilingual model released on December 6, 2024, built on an optimized transformer with grouped-query attention and a 128K-token context. ...
Model lineup
Context window and positioning from each provider.
| Model | Context | Positioning |
|---|---|---|
| DeepSeek V4 Pro | 1049K | DeepSeek V4 Pro is the large tier of DeepSeek's V4 family, a 1.6T-parameter mixture-of-experts model with 49B active parameters, introduced as an… |
| Kimi K3 | 1049K | Kimi K3 is Moonshot AI's flagship model, a 2.8-trillion-parameter open-weight mixture-of-experts model with native visual understanding and a… |
| GLM 5.2 | 1049K | GLM-5.2 is Z.ai's open-weight flagship text model released in June 2026 for long-horizon tasks, and the first GLM model to deliver that capability… |
| DeepSeek V3.2 | 131K | DeepSeek-V3.2 is DeepSeek's open-weight, reasoning-first model released in December 2025 as the official successor to V3.2-Exp |
| MiniMax M3 | 1000K | MiniMax-M3 is MiniMax's flagship multimodal foundation model, which MiniMax describes as open-weight |
| MiMo-V2.6-Pro | 1050K | MiMo-V2.6-Pro is the flagship of Xiaomi's open-source MiMo-V2.6 series, a trillion-parameter, natively omni-modal reasoning model released in… |
| MiMo-V2.6-Flash | 1050K | MiMo-V2.6-Flash is the efficiency tier of Xiaomi's open-source MiMo-V2.6 series, released in September 2026 as a natively omni-modal reasoning model… |
| Llama 3.3 70B Instruct | 131K | Llama 3.3 70B Instruct is Meta's instruction-tuned, text-only multilingual model released on December 6, 2024, built on an optimized transformer… |
| Qwen3.5-35B-A3B | 262K | Qwen3.5 35B A3B is an open-weight native vision-language model from Alibaba's Qwen3.5 series, released in February 2026, pairing linear attention… |
| Qwen3.6 27B | 262K | Qwen3.6 27B is an open-weight dense 27-billion-parameter vision-language model in Alibaba's Qwen3.6 family, released in April 2026 with thinking and… |
| gpt-oss-120b | 131K | gpt-oss-120b is OpenAI's largest open-weight reasoning model, released in August 2025 under the Apache 2.0 license: a 117B-parameter… |
| Qwen3.8 2.4T A95B | 1000K | Qwen3.8 2.4T A95B is the open-weight release of Alibaba's Qwen3.8 Max-class model, a sparse Mixture-of-Experts with 2.4 trillion total and about 95… |
Open models, served for you
Open-weight models can be self-hosted, fine-tuned and audited. SandBase serves them over the OpenAI-compatible POST /v1/chat/completions, so you can prototype and run production traffic without managing GPUs, then self-host later if you need to.
Every model on this page is described as open-weight or open-source in its catalog entry.
On this page
- Large MoE models: DeepSeek V4 Pro and V3.2, Qwen3.8 2.4T A95B, Kimi K3.1 Pro and K3, GLM 5.2, MiniMax M3, MiMo V2.6 Pro, Hunyuan Hy4 Preview.
- Compact and efficient models: Qwen3.8 27B, Qwen3.6 27B, Qwen3.5 35B A3B, MiniMax M3.1 Flash, MiMo V2.6 Flash.
- Western open models: OpenAI gpt-oss-120b and Meta Llama 3.3 70B.
Licences differ by model; check the provider's licence before you self-host or fine-tune. More Chinese models are on Chinese LLM APIs.
Start using Open-Source LLM APIs on SandBase
Switch models by changing one field.
- 01
Create an API key
Sign in to the SandBase Console and create a key on the API Keys page. Keep it in an environment variable and never commit it to source control.
TERMINALexport SANDBASE_API_KEY="YOUR_SANDBASE_API_KEY" - 02
Send a chat request
Call the OpenAI-compatible Chat Completions API with model deepseek/deepseek-v4-pro. Any OpenAI SDK works with base URL https://api.sandbase.ai/v1.
TERMINALcurl https://api.sandbase.ai/v1/chat/completions \ -H "Authorization: Bearer $SANDBASE_API_KEY" \ -H "Content-Type: application/json" \ -d '{"model": "deepseek/deepseek-v4-pro", "messages": [{"role": "user", "content": "Explain vector databases in two sentences."}]}' - 03
Connect your AI tools
Use the same key in Claude Code, Codex, Cursor and other clients. See Use in AI tools for setup.
Questions, answered
Why use an API for an open model?
To avoid running GPUs, scaling and upgrades yourself. Your prompts and outputs stay portable if you move to self-hosting later.
Are these models free?
The weights are openly available, but hosted inference is billed per token. Prices are on each card.
Can I move to self-hosting later?
Yes. Requests use the OpenAI-compatible format, which most self-hosted inference servers also accept.
Which licence applies?
Each model’s own licence from its provider. Check the provider page before commercial self-hosting.
Build with Open-Source LLM APIs.
One key, consistent APIs, and usage-based pricing across every model on this page.