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.
DeepSeek V4 Flash Vision Exp is an experimental vision-enabled build of V4 Flash, and it takes an unusual approach to image cost: every image is capped at 384 tokens regardless of its resolution. A two-megapixel scan and a twenty-five-megapixel one bill identically, because both are resized to the same working size before inference. That turns document processing from an open-ended cost into arithmetic you can do before sending the request. It accepts up to six hundred images per call, matches the text model it is built on across agents, reasoning, and world knowledge, and holds the same million-token context. On DEVUP AI it is the choice when the workload is pages rather than paragraphs.
GLM-5.3 is Z.ai's flagship coding model, and it makes an unusual argument: it shares its base model entirely with the previous generation. No new pre-training, no larger foundation — every gain comes from post-training alone, and the gains are large. Terminal Bench 3.0 moves from 4.6 to 28.3 on identical weights. Alongside coding, Z.ai reports that cyber capability emerged faster than expected during that scaling, putting the model at the state of the art for vulnerability discovery and more than doubling its predecessor further up the exploitation chain. It holds a one-million-token context with a 128K maximum output, generates roughly three times faster than comparable frontier models, and reasons at three selectable effort levels.
GLM-5.3-Flash is the first natively multimodal model in the GLM-5 series, and it was built from a new base model rather than adapted from an existing one. It carries 320B total parameters but activates only 18B per token, routing each token through 8 of 288 experts across 45 layers. Its defining architectural choice is a hybrid attention stack combining linear and sparse attention — a first for the series — which cuts the serving cost of very long inputs while keeping long-context accuracy intact. Trained on a 30-trillion-token multimodal corpus, it accepts images alongside text and holds a full one-million-token context. Released under the MIT license, it is the model to reach for on DEVUP AI when your input is visual, very long, or both.
Granite 4.2 30B is the flagship of IBM's Granite 4.2 reasoning family and the model that goes furthest through its training pipeline. On top of the chain-of-thought reasoning and the agentic reinforcement-learning block shared with the mid-size tier, the 30B adds a third round of verifiable-reward training and a second supervised fine-tuning phase dedicated entirely to agentic coding — reading real repositories, editing files, and running test suites. It leads the family across software engineering, terminal operation, mathematics, and long-context recall, holding a 128K window with the strongest retention of the three sizes. Tested across twelve languages and released under Apache 2.0, it is the model to reach for on DEVUP AI when the task is genuinely hard and a wrong answer costs more than a slow one.
Granite 4.2 8B is IBM's mid-size open reasoning model, and the smallest one in its family trained to act. Beyond the chain-of-thought reasoning shared across Granite 4.2, the 8B goes through an additional agentic reinforcement-learning stage in real sandboxed environments — editing real repositories, driving a live terminal, and running multi-hop web searches — which is why it is the smallest Granite 4.2 model with reported agentic coding results. Three thinking modes let you trade reasoning depth against latency per request, and it holds a 128K context with substantially better long-context recall than the compact tier. Tested across twelve languages and released under Apache 2.0, it is the balanced default on DEVUP AI for production agents, coding assistants, and long-document work.
Granite 4.2 3B is IBM's compact open reasoning model — a three-billion-parameter dense transformer that thinks before it answers. It emits an explicit chain of thought inside <think> tags before producing a final response, which lifts its accuracy on math, code and multi-step logic well beyond what its size would suggest. Three thinking modes — full, low-effort, and off — let you trade depth against latency on a per-request basis, so a single model can serve both a fast autocomplete and a careful agent. It handles a 128K context window, calls tools with explicit reasoning about why each call is made, and is tested across twelve languages. Released under Apache 2.0, it is a strong default on DEVUP AI for agentic workflows, structured extraction, and reasoning tasks that need to stay lightweight.
L3-8B-Lunaris-v1 is an 8-billion parameter large language model developed by Sao10K, based on the Meta Llama 3 architecture. It is a strategic 5-way model merge designed to serve as a versatile generalist and an immersive roleplaying assistant. Created as an improvement over the Stheno v3.2 series, Lunaris balances high creative output with enhanced logical consistency and reasoning capabilities, making it highly effective for both imaginative storytelling and structured conversational tasks.
L3.1-70B-Euryale-v2.2 is a 70-billion-parameter large language model fine-tuned by developer Sao10K on top of the Meta Llama 3.1 70B Instruct foundation. Specifically optimized for advanced creative writing, storytelling, and highly immersive interactive roleplay, it features a massive 131K context window for maintaining long-term memory and plot coherence. Trained on multi-turn conversational datasets synthesized from Claude 3.5 Sonnet, this open-weights model stands out for its deep spatial awareness, strict compliance with complex system prompts, and uncensored flexibility, making it a premier choice for generating rich, uninhibited narrative prose.
The gemma-4-E4B-it model is a lightweight, open-weights vision-language-audio model developed by Google DeepMind. It is optimized for high-performance, private, and offline on-device execution on smartphones, laptops, and IoT hardware. Featuring 4.5 billion effective parameters (8 billion total with embeddings) built on an innovative Per-Layer Embedding architecture, this instruction-tuned ("it") variant delivers state-of-the-art multimodal reasoning across text, images, and native audio. It also boasts a massive 128K context window and a built-in reasoning "thinking mode" to eliminate logic errors during local execution.
Qwen3.5-122B-A10B is a large Mixture-of-Experts model from Alibaba's Qwen3.5 series with 122B total parameters and 10B activated per token. It features a 262K token context window (extensible to 1M with YaRN), thinking/reasoning mode, tool calling, and support for 201 languages. Excels at complex reasoning, coding, multimodal understanding, and agentic tasks with the efficiency of sparse activation.
The MiniMax-M2.7 model is a next-generation, high-speed text model built on a sparse Mixture-of-Experts (MoE) architecture. It stands out as a pioneering Self-Evolving AI, utilizing continuous recursive self-improvement pipelines during development. Engineered explicitly for advanced software engineering and complex agentic workflows, it delivers exceptional generation speeds and high-fidelity output for enterprise multi-agent networks, debugging, and productivity automation.
Qwen3.8-27B is the compact member of Alibaba's most capable open model generation, and it makes an unusual argument: that a dense 27-billion-parameter model can compete with systems a hundred times its size on the work most applications actually do. It is a native vision-language model reading images and hour-scale video, with a hybrid attention stack that mixes gated linear attention with full attention across 64 layers, and a 262K native context extensible far beyond it. Thinking is controllable along three separate axes — on or off, three depth levels, and whether reasoning carries forward across turns. It leads its class on document intelligence and structured extraction, and on DEVUP AI it is the choice when the work is high-volume and the documents are real.
