EXPLORE / 01
Models index
Foundation models, releases, and capability boundaries
RECORDS
Entity records
35 records · 1/1
- 001
Qwen-Image-2.1
An open-weight unified multimodal image generation and editing model released by Alibaba's Qwen team in September 2026. Built on a 7B-parameter 32-layer Single-Stream DiT backbone and a Qwen3-VL 8B encoder, it natively supports 2K resolution, RGBA transparent image generation, and multi-reference editing with up to 10 condition images while leveraging prefix KV cache reuse for inference efficiency.
qwen.ai - 002
Jev
A non-generative System 1 decision model developed by TypeSafe AI that directly outputs type-safe primitives (Noul, Choice, and Score) instead of text strings, operating with 70ms-500ms latency and calibrated probabilities trained via RLCD.
typesafe.ai - 003
GLM-5.3-FlashX
A high-speed inference large language model by Zhipu AI built on a 320B total / 18B active parameter hybrid attention architecture with a 1M context window and generation speeds up to 200 tokens/s.
bigmodel.cn - 004
Qwen3.8-Omni-Flash
A lightweight native omnimodal model released by Alibaba's Qwen team in September 2026, supporting text, image, audio, and video inputs with a 1M-token context window, optimized for agentic task planning and tool execution in production environments.
qwen.ai - 005
GPT-6 Astra
OpenAI's frontier flagship model featuring a 1.05M-token context window and 128k output, focused on autonomous computer operation and long-horizon tasks while being the first system to reach the Critical cybersecurity risk threshold under its Preparedness Framework.
openai.com - 006
Claude Fable 5.1
Anthropic's frontier flagship model released in September 2026, featuring a 1M-token context window and 128k output limit, engineered for long-horizon autonomous coding, multi-file refactoring, and adaptive deep reasoning.
anthropic.com - 007
Gemini 3.8 Flash
Google's efficient workhorse multimodal model featuring a 1M-token context window and 64k output limit, tuned for long-horizon software engineering, terminal execution, and autonomous agents.
deepmind.google - 008
Qwen3.8-27B
An open-weights dense 27B vision-language model open-sourced by Alibaba Cloud's Qwen team. Built on the Qwen3.5 architecture, it features a 64-layer hybrid attention design (48 linear and 16 gated), native MTP draft head, and thinking mode, supporting native 262K context (scalable to 1M tokens) for single-GPU local deployment and long-horizon agentic tasks.
huggingface.co - 009
GLM-5.3
A 743B-parameter flagship model released by Zhipu AI, enhanced through post-training reinforcement learning scaling with IndexShare, SAO, and Slime for complex terminal execution, coding, and cybersecurity evaluation.
z.ai - 010
Gemini 3.7 Flash
Google's efficient workhorse multimodal model featuring a 1M-token context window and tunable thinking, optimized for long-horizon coding, complex reasoning, and agentic workflows.
blog.google - 011
Qwen3.8-Max
A flagship multimodal foundation model by Alibaba, built on a 2.4-trillion-parameter sparse MoE architecture with 95 billion activated parameters, supporting 1-million-token context and long-horizon autonomous task execution.
qwen.ai - 012
Seedance 2.5
A next-generation multimodal audio-visual generation model by ByteDance, featuring 30-second single-clip generation, 50-slot multimodal reference conditioning, and timestamp-based local editing.
bytedance.com - 013
DeepSeek-V4-Flash
An efficiency-optimized Mixture-of-Experts (MoE) model by DeepSeek, featuring 284B total and 13B active parameters, supporting a 1M context window, and tailored for high throughput, low latency, and agentic workflows.
deepseek.com - 014
Claude Opus 5
Anthropic's frontier reasoning model approaching Claude Fable 5 intelligence, featuring dynamic effort settings and default thinking.
anthropic.com - 015
Gemini 3.5 Flash Cyber
A domain-specific model fine-tuned by Google for cybersecurity, integrated into CodeMender for multi-agent automated vulnerability discovery and remediation.
blog.google - 016
Gemini 3.5 Flash-Lite
A high-throughput, low-latency model released by Google in July 2026, reaching 350 output tokens/sec with built-in computer use tools and configurable thinking levels.
blog.google - 017
Gemini 3.6 Flash
An upgraded workhorse model released by Google in July 2026. It reduces output token usage by 17% compared to Gemini 3.5 Flash while boosting DeepSWE coding benchmarks by up to 65%, optimized for agentic tool-calling.
blog.google - 018
Qwen-Image-3.0
A multimodal image generation base model released by Alibaba's Tongyi Lab in July 2026. Focusing on real-world productivity, it features strong long-text comprehension, complex layout control, precise micro-text rendering, and multilingual UI simulation.
qwen.ai - 019
Qwen-Audio-3.0-TTS
A hosted, production-oriented text-to-speech (TTS) system released by Alibaba's Tongyi Lab in July 2026. It offers a Plus tier for high-quality dubbing and a Flash tier for ultra-low latency real-time voice agents, topping independent quality leaderboards.
bailian.console.aliyun.com - 020
Qwen3.8-Max-Preview
Alibaba Qwen's 2.4-trillion-parameter flagship preview model, first available through Qwen Cloud Token Plan, Qoder, and QoderWork for coding, data analysis, Office workflows, and complex long-horizon agentic tasks. Qwen has previewed an open-weight release, while the date, license, and full evaluations remain undisclosed.
qwen.ai - 021
Kimi K3
Moonshot AI's flagship multimodal reasoning model released in July 2026. It has 2.8 trillion total parameters and combines an MoE architecture with Kimi Delta Attention and Attention Residuals, native vision, and a 1-million-token context window for long-horizon coding, knowledge work, and deep reasoning.
kimi.com - 022
Qwen-Audio-3.0-Realtime
Alibaba's real-time voice interaction model featuring millisecond-level low latency, full-duplex interaction, and proactive Agent tool execution. It is available in Plus (reasoning optimized) and Flash (speed optimized) versions.
qwen.ai - 023
Gemini 3.5 Pro
Google's flagship large language model expected to launch in July 2026, featuring a 2-million-token context window, deep retraining, and optimization for complex logical reasoning and coding tasks.
deepmind.google - 024
Kimi K2.7 Code
Large language model provided by Moonshot AI.
moonshot.cn - 025
Gemini 3.5 Flash
Large language model provided by Google.
deepmind.google - 026
GLM 5.2
Large language model provided by Zhipu AI.
zhipuai.cn - 027
GPT-5.4
Large language model provided by OpenAI.
openai.com - 028
GPT-5.5
Large language model provided by OpenAI.
openai.com - 029
Claude Opus 4.8
Large language model provided by Anthropic.
anthropic.com - 030
Claude Fable 5
Large language model provided by Anthropic.
anthropic.com - 031
GPT-5.6 Luna
Large language model provided by OpenAI.
openai.com - 032
GPT-5.6 Terra
Large language model provided by OpenAI.
openai.com - 033
GPT-5.6 Sol
Large language model provided by OpenAI.
openai.com - 034
Hy3
**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.
Website pending - 035
Claude Sonnet 5
Large language model provided by Anthropic.
anthropic.com