盘点近两月新模型，含 OpenAI Dots

0xHades @OnlyHades_原推发布于 2026.10.02

在 X 上打开原推

The AI model releases over the last ~2 months have been pretty crazy. We really are in the Super Intelligence era. Which ones have been most useful to you? ➜ Claude Opus 5.5 by @AnthropicAI - advanced coding, software engineering, research and long-running agent tasks. - built to work through large codebases and complex multi-step projects. ➜ Claude Sonnet 5.5 by @AnthropicAI - faster, cheaper model for coding, analysis, customer support and everyday agent workloads. - basically the high-volume version of Claude 5.5. ➜ Claude Fable 5.1 by @AnthropicAI - deeper reasoning for research, coding and difficult knowledge-work tasks. - designed for jobs where you want more deliberate reasoning rather than quick responses. ➜ Claude Mythos 5.1 by @AnthropicAI - specialized version focused on high-risk domains like cybersecurity and life sciences, with tighter access controls. ➜ GPT-6 Astra by @OpenAI - general-purpose frontier model with strong computer use, coding, browsing, science and professional-work capabilities. - designed to operate tools and complete multi-step tasks. ➜ GPT-6 Sol by @OpenAI - reasoning model for coding, analysis and agentic workflows at lower cost than Astra. - built for developers who need strong reasoning at higher volume. ➜ GPT-6 Luna by @OpenAI - lightweight, low-cost model for high-volume inference. - useful for classification, extraction, simple reasoning and applications that need millions of model calls. ➜ GPT-6.1 Sol by @OpenAI - upgraded Sol with stronger reasoning, coding, computer use and multi-agent capabilities. ➜ Dots by @OpenAI - persistent personal agent that can handle tasks across different applications instead of simply answering prompts in a chat window. ➜ ChatGPT Images 2.5 by @OpenAI - image generation and editing with better consistency, instruction following and control over complex scenes. ➜ Gemini 3.8 Flash by @GoogleDeepMind - fast multimodal model for text, images, coding, reasoning and agent tasks. - built for applications where latency and cost matter. ➜ Gemini 3.8 Live by @GoogleDeepMind - real-time multimodal conversation. - it can listen, see visual input and respond naturally while reasoning during the interaction. ➜ Gemini 3.8 Flash TTS by @GoogleDeepMind - turns text into natural speech with control over voice, pacing and conversational delivery. - built for voice agents and applications. ➜ Gemini 4 Argon by @GoogleDeepMind - long-context reasoning model aimed at coding, research, finance, legal work and cybersecurity. - its huge context window lets it work across very large amounts of information in one session. ➜ Grok 4.7 by @xai - reasoning and coding model designed for long-running tasks. - strong focus on software engineering, research and tasks that require the model to repeatedly check and improve its own work. ➜ Jev by @typesafeai - decision model rather than a traditional chatbot. - takes structured information and outputs probabilities or decisions, making it useful for routing, fraud detection, classification and agent tool selection. ➜ Muse Spark 1.3 by @AIatMeta - coding and multimodal agent model that can plan tasks, use tools, maintain context and work through longer workflows. ➜ DeepSeek V4 Pro by @deepseek_ai - open-weight model for reasoning, coding, research and agentic workflows. - designed to give developers frontier-level capabilities without relying entirely on closed APIs. ➜ DeepSeek V4.1 Flash by @deepseek_ai - cheaper high-throughput reasoning model with a huge context window. - designed for coding, tool use and running agents at scale. ➜ MiMo-V2.6 Pro by @XiaomiMiMo - large open-weight reasoning model for coding, research, math, cybersecurity and long-horizon tasks. ➜ MiMo-V2.6 Flash by @XiaomiMiMo - smaller/faster version designed for high-volume inference while keeping strong reasoning and coding capabilities. ➜ Qwen3.8 Max by @Alibaba_Qwen - large multimodal model for coding, reasoning, research and long-context tasks. - built to handle complex workflows rather than just standard chat. ➜ Qwen3.8 Omni-Flash by @Alibaba_Qwen - multimodal model that can understand text, images, audio and video in one system, with support for real-time interaction and agent workflows. ➜ Qwen-Image 2.1 by @Alibaba_Qwen - image generation and editing model focused on text rendering, visual consistency and detailed image creation. ➜ GLM 5.3 by @Zai_org - open-weight reasoning model focused on coding, research, mathematics and long-horizon agent tasks. ➜ GLM 5.3 Flash by @Zai_org - faster, cheaper version designed for high-volume agent and coding workloads. ➜ Griffin by @tavus - real-time video-to-video model that can see, hear and respond during a live video conversation. - built specifically for human-like video agents. ➜ SWE-2 by @cognition - coding model optimized specifically for autonomous software engineering agents. - it can understand repositories, write code, debug and work through multi-step engineering tasks. All I see is more practical autonomous agents which is net bullish for crypto. An agent that can think is useful. An agent that can think + hold money + make payments + interact with markets + pay other agents + execute transactions is an economic actor. This means more agents will eventually need to hold money, make payments, trade and transact onchain. Greater times ahead.