large language model

Bonsai 27B

The first 27B-class multimodal large model that runs locally on a smartphone, compressed to 3.9GB from Alibaba Qwen3.6 27B via extreme quantization

Kimi K3 Open-Source Edition

Moonshot AI's largest open-source LLM in the world: 2.8 trillion parameters, native multimodality and a 1M-token context window, topping global coding benchmarks

Zhipu GLM-5.2

Zhipu's new flagship LLM, launched and open-sourced under the MIT license: competes with top closed-source models on coding and long-horizon tasks and is adapted to domestic Chinese compute

Osaurus

A macOS native open source AI runtime that runs AI models and manages agents locally, keeping data and privacy entirely on your Mac.

Gemini 3.6 Flash

Google DeepMind is an efficient main model for the era of intelligent agents. It is stronger and more token-saving in programming, knowledge work and multi-modal tasks.

Panshi·Science Basics Large Model 2.0

The large-scale integrated scientific model created by the Chinese Academy of Sciences provides a highly reliable and interpretable intelligent base for scientific research.

SenseNova U1 Pro

The flagship native multi-modal large model released by SenseTime, focusing on delivery-level complex graphic and text creation and native 8K ultra-clear output

Qwen3.8

Alibaba Tongyi Qianwen's 2.4 trillion parameter new generation flagship multi-modal large model, focusing on code engineering and professional office, the preview version has been launched and the open source weight is promised

Muse Spark 1.1

The flagship multi-modal reasoning model launched by Meta Super Intelligence Laboratory, specially built for Agent tasks and programming, with millions of token contexts, and the price is only a quarter of competing products at the same level.

AI21 Jamba2

The open source hybrid SSM-Transformer model family launched by Israel's AI21 Labs focuses on enterprise-level reliability and memory efficiency of 256K long context.

Gemini 5

Gemini 3.5 Pro, the flagship large model released by Google DeepMind in July 2026, features 2 million token ultra-long context and Deep Think extended reasoning

GPT-7

OpenAI’s current strongest cutting-edge model family, GPT-5.6 (Sol/Terra/Luna), reshapes complex task execution with agent workflow and native tool calls

Claude Opus 6

Anthropic Opus flagship frontier, focusing on long-range reasoning, long-term autonomous agent and "honest" self-calibration

Grok 6

xAI will start a price war with Grok 4.5 (1.5 trillion parameters, focusing on programming and Agent) in the second half of 2026, and Grok 4.6 (2 trillion parameters) is expected to take over in August

Upstage Solar 2

Solar Pro 2, the second-generation flagship large language model launched by South Korea's Upstage, has 31 billion parameters, can be run on a single card, dominates the Korean language list, and has solid enterprise-level implementation.

Fireworks 2

An open source model inference and training cloud platform founded by the original PyTorch team, focusing on customizing "exclusive intelligence" on enterprise private data

DeepSeek V4.5

In-depth exploration of the iterative version of the flagship large model launched in July 2026, strengthening code, Chinese long text and multi-round consistency, continuing the open source and floor-price play style

Claude Opus 5

Anthropic's new-generation Opus flagship model launched in July 2026 features million-level context and super agentic coding. It is officially called the world's strongest programming model, approaching the capabilities of Fable 5 at about half the price.

Gemini 4.5 Pro

Google's high-performance flagship model featuring millions of tokens, long context and full-modal understanding

Aleph Alpha Columbus

Aleph Alpha 面向欧洲 enterprises and 政府的主权 AI platform，以合规, 数据可控 and 可解释性为核心

NetEase Yuyan 2

网易伏羲自研的中文角色扮演large model，专攻剧情generate and 人设对话，2026 年 6 月through 网信办备案

SenseTime SenseNova 6

商汤第六代原生multimodal large model家族，以统一架构enables 文本, 图像, 视频, 语音的深度推理，并延伸出 6.7 Flash-Lite 办公智能体 and open-source U1 统一模型

iFlyTek Spark 6

科大讯飞based on 全国产算力训练的认知large model，语音 and 教育两大场景极强，政企落地领先，但通用对话 and 代码能力偏中下

Baidu ERNIE 5.5

百度旗下全能 AI assistant，深耕中文知识问答, 文学创作 and 逻辑推理场景，具备视频, 图片, 语音等multimodal AI 能力

Qwen3.5-Coder

Alibaba Tongyi Qianwen Qwen 3.5 family's code-specific open weight model, 7B/32B dual size, 256K context, Apache 2.0, commercially available, focusing on locally deployed warehouse-level programming and Coding Agent

Grok 5.5

xAI’s real-time AI assistant, the only native access to X real-time data stream, the Grok 5 beta version is the 6 trillion parameter MoE flagship

Nemotron Ultra 253B

NVIDIA 253B open weighted reasoning model, closed source flagship for text reasoning and coding benchmarking, available for commercial use

Gemma 5

The latest generation of Google DeepMind open weight model family Gemma 4 (Apache 2.0, covering mobile phones to servers)

Claude Haiku 5

Anthropic's fastest and cheapest model (currently Haiku 4.5), with coding capabilities close to flagships to minimize cost and latency

Mistral Codestral 2

Mistral's second-generation code-specific model, Apache 2.0 can be legally self-hosted and commercially used after re-authorization, and is good at IDE inline completion