Models

U2 Flash focuses on Agent and Coding scenarios, with strong complex reasoning, code generation, and task execution. It supports long context for complex engineering, long-horizon tasks, and multi-step Agent workflows—delivering faster inference and better cost efficiency as a high-performance, cost-effective foundation for developers and enterprises.

U2 Decision is Unisound’s open-sourced enterprise decision model (4B). As the decision layer across multi-model, multi-agent, and multi-tool collaboration, it focuses on who should run a task, whether it can run, and which path it should take—via intelligent model routing, safety decisions, and workflow routing.

Kimi's strongest flagship with 2.8T parameters, KDA hybrid linear attention, native vision, and 1M-token context for long-horizon coding, knowledge work, and reasoning.

GLM-5.3 is Zhipu's strongest coding model to date, about 50% stronger than GLM-5.2 on internal evals; on white-box code review and vulnerability discovery it matches Mythos 5, showing strong potential for cybersecurity defense.

GLM-5.3-Flash is Zhipu's first native multimodal GLM-5 model, delivering stronger intelligence than GLM-5.2 at ultra-low cost. Built on a 320B / 18B active sparse + linear-attention hybrid architecture with vision natively integrated into the coding loop, it can observe UIs, render results, and coordinate across code, browsers, and GUIs — and extend to Office, finance research, and professional documents with autonomous planning and tool use.

Zhipu flagship for long-horizon tasks with 1M lossless context. Full-stack engineering from design to deployment for complex projects.

High-value Plus model with upgraded vision-language, coding, tools, and agent workflows. Scene perception, GUI ops, and visual-reference code gen.

Qwen3.8-Flash is Qwen's latest multimodal model with strong understanding, generation, and speed. Native million-token context for long docs, repos, and complex chats; strong at coding assistance, agent collaboration, and vision; OpenAI/Anthropic compatible for tools like Claude Code and Codex.

2.4T-parameter MoE flagship with major coding and office gains; native vision across planning, execution, and verification; deep parsing of long docs and videos.

Unisound's flagship U2 model with high intelligence per token and native agent capabilities. Compact yet capable, lower inference cost; autonomously plans, executes, and refines tasks end to end.

U2-Med has deep expertise in medical knowledge, healthcare insurance policies, pharmaceutical rules, and medical literature. It accurately parses diverse professional content including medical records, examination reports, prescriptions, policy clauses, and academic literature. Whether for everyday users interpreting health check reports and consulting on medication and insurance reimbursement, or for healthcare institutions conducting insurance audits, claims processing, drug development, and clinical research, it delivers precise, efficient, and trustworthy intelligent support.

U2-RadiMed is Unisound's medical imaging LLM for image+text input. It helps doctors read scans, locate lesions, aid diagnosis, and auto-generate structured reports — an end-to-end AI assistant from image analysis to clinical decision support.

Human-like TTS with style, emotion, and multi-language/dialect support. Async long-text synthesis and flexible audio output for production scale.

Clone voices from minimal samples with natural, emotional speech. Persistent cloned voices for reusable brand assets.

U2-TTS-Design does not require uploading real-person samples or picking from a fixed voice library. Simply describe the voice you want in natural language (e.g. gender, age, pitch, speed, emotion, accent, texture, character personality, use case), and it can generate a reusable exclusive voice from scratch.

Universal ASR with context understanding and custom vocabulary. Multi-dialect, bilingual, with one-shot and real-time transcription.

Intelligent document parsing beyond OCR—classification, layout restoration, and key info extraction. Converts unstructured docs to structured data.

Medical document intelligence for records, reports, prescriptions, and bills. Handles handwriting, abbreviations, stamps, and complex layouts.

Efficient MoE (284B total, 13B active) with million-token context. Fast, low-latency, cost-effective for chat, content, RAG, and batch copywriting.

Flagship MoE (1.6T total, 49B active) with million-token context. Top math, reasoning, code, and long-text analysis for research and advanced agents.

Multimodal agent model unifying vision-language understanding, instant/thinking modes, and chat/agent paradigms.

Kimi K2.6 offers stronger, more stable long-horizon coding, better instruction following and self-correction. Supports text, image, and video input; thinking and non-thinking modes; chat and Agent tasks.

Long-horizon model (744B params, 200K context, 128K output). Strong reasoning, long-text, and code generation for interactive and enterprise use.

Next-gen model for coding and agents—open-source SOTA in complex engineering and long tasks, nearing Claude Opus in real coding.

MiniMax M3 delivers leading coding and agentic ability, a 1M context window, and native multimodality for long documents, content generation, coding, bugfix, and app building, with end-to-end agentic workflows and natural mixed media interaction.

Production-grade agent model built for real-world productivity—faster, stronger, and more cost-effective.

Largest Max text model with broad, deep agent skills—coding, productivity, and long autonomous runs.

Native vision-language Plus rivaling frontier models. Major gains over 3.5 in agentic coding, frontend, OCR, and object localization.

Fast vision-language model with major gains over 3.5-Flash. Stronger agentic coding, math, code reasoning, and spatial detection.

35B-A3B vision-language MoE with linear attention and sparse experts. Better efficiency; gains in agentic coding, reasoning, and spatial detection.