Model Library
Browse and deploy state-of-the-art AI models through the DEVUP Gateway.
Browse and deploy state-of-the-art AI models through the DEVUP Gateway.
Browse and deploy state-of-the-art AI models through the DEVUP Gateway.
Qwen3.8-2.4T-A95B is the first Qwen-Max-class model released as open weights: a sparse Mixture-of-Experts design with 2.4 trillion total parameters, of which 95 billion activate per token, routing through 10 of 512 experts plus a shared one. Its 92-layer backbone alternates gated linear attention with full attention every fourth block, which is what keeps inference cost near that of a far smaller model while sustaining a 262K context. It is built for the hardest end of the workload spectrum — long-horizon agentic execution, professional research, and repository-scale engineering — and it always thinks before it answers, with three tunable depth levels. Text in, text out, with no mode that skips deliberation.
Muse Glimmer 30B is Meta Superintelligence Lab's first open-weight release, and it was designed around a single constraint: an agent that runs on the machine in front of you. Roughly 29.6 billion dense parameters including a dedicated vision encoder, distilled from a larger model, and quantized to fit under 20 GB — small enough to run on one consumer GPU while still holding a 131K context and reading screenshots, charts, and documents. What sets it apart is what it was trained for rather than what it knows: sustained tool use across long workflows, and diagnosing a failed tool call and retrying rather than stopping. It ships with published prompt-injection and privacy evaluations, which few models of any size do. Apache 2.0.
Kimi K3 is Moonshot AI's most capable model and the first open model in the 3-trillion-parameter class. It carries 2.8T total parameters but activates only 104B per token, routing through 16 of 896 experts, and it is built on a new attention design that mixes Kimi Delta Attention with gated latent attention across 93 layers. Vision is native rather than bolted on, with a dedicated encoder inside the model, and the context window runs to a full million tokens. It was built for work that takes a long time to finish: sustained engineering sessions across large repositories, terminal orchestration, deep research, and document-heavy knowledge work. Thinking is always on, with three effort levels, and the model returns its reasoning as a first-class field.
Qwen3.8-Max is Alibaba's flagship model and the complete form of the Qwen3.8 generation: a Mixture-of-Experts design with 2.4 trillion total parameters, 95 billion of which activate per token, running against a one-million-token context window with native text, image, and video input in a single endpoint. It reads PDFs of two hundred pages and more, and it uses vision throughout execution rather than only at the start — inspecting its own intermediate output, catching a misaligned interface or a wrong-facing object, and revising the plan. Thinking is hybrid: on by default, switchable off per request, and tunable across three depth levels, with reasoning traces that carry across turns so long agent loops keep their chain of thought.
The model prioritizes token efficiency and agentic inference at production scale, stretching what developers can achieve within limited token, latency, and serving-cost budgets.
Inkling-Small is a Mixture-of-Experts transformer with 276B total parameters, 12B active, trained on NVIDIA GB300 NVL72 systems. Like Inkling, it features native reasoning over audio and images, variable thinking effort
The google/gemma-4-31B-it-Ultra is a highly capable, multimodal, dense open-weights model developed by Google DeepMind. Featuring 30.7 billion parameters and a massive 256,000-token context window, it excels in advanced reasoning, complex coding, and multimodal tasks (including text, high-resolution images, and video temporal reasoning). Released under the permissive Apache 2.0 license, this "Ultra" designation highlights its optimization for autonomous agent workflows, precise function calling, and structured data generation. When quantized (e.g., 4-bit), it can be comfortably run locally on consumer-grade hardware like a 24GB RTX 4090.
The openai/gpt-oss-120b is a highly efficient, open-weights Mixture-of-Experts (MoE) language model released by OpenAI under the permissive Apache 2.0 license. While it features 117 billion total parameters, it activates only 5.1 billion parameters per token, allowing it to run entirely on a single 80GB GPU. Designed specifically for advanced AI agents, it features configurable reasoning efforts (low, medium, high), transparent Chain-of-Thought (CoT), and native optimization for tool use and structured data generation.
Claude Opus 5 is Anthropic's most advanced Opus model, powering long-running agents while delivering improvements in coding and professional work.
Inkling is a general-purpose multimodal model that accepts text, image and audio inputs and generates text outputs.
Hy3 is a 295B-parameter Mixture-of-Experts (MoE) model with 21B active parameters and 3.8B MTP layer parameters, developed by the Tencent Hy Team. Following the Hy3 Preview launch in late April, we gathered feedback from 50+ products and scaled up post-training with higher quality data. Today, we introduce Hy3, which outperforms similar-size models and rivals flagship open-source models with 2-5x parameters. It also shows significant gains in utility across various products and productivity tasks.
Claude Fable 5 is Anthropic's next generation of intelligence for the hardest knowledge work and coding problems. It works independently for longer than any prior generally available Claude model: run it in an agent harness and it can work for days at a time, planning across stages, delegating to sub-agents, and checking its own work.
