AI UP LAB面向任务的 AI 实践资源AI resources for real tasks
收藏Saved
更多More
StarStar

模型与能力Models & capabilities

了解模型的能力、获取方式和适用边界。官方公告、第三方评测和社区展示分开标注，效果主张请对照来源。Understand capabilities, access, and limits. Official announcements, independent evaluations, and community demos are labelled separately; check claims against their sources.

共 28 个模型 · 精选在前28 models · Featured first

Anthropic Anthropic 官方 Official 社区 showcase Community showcase + 精选 + Featured

Claude Haiku 5.5：Anthropic 最便宜、最快的小模型，价格约降四分之三 Claude Haiku 5.5: Anthropic's cheapest, fastest small model, about 75% cheaper to run

Claude Haiku 5.5: Anthropic's cheapest, fastest small model, about 75% cheaper to run

Anthropic 2026-10-07 发布 Claude Haiku 5.5，定位是量大、怕贵、要快的活：摘要、分类、数据库查询、在线客服、操作浏览器，或者给 Opus 5.5、Sonnet 5.5 当写代码时的小帮手。官方说平均跑起来比 Haiku 4.5 便宜约 75%。价格按每百万 token 算，提示词在 10 万 token 以内时输入 0.10 美元、输出 0.50 美元，超过 10 万时是 0.50 和 2.50。这是第一款能调「思考力度」的 Haiku。同一天 Sonnet 5.5 的缓存读取降价一半，Max 和 Team 订阅也开始送每月 API 额度。平台模型名 claude-haiku-5-5，各大云平台都能用。 Anthropic released Claude Haiku 5.5 on 2026-10-07 for high-volume, cost- and speed-sensitive work: summaries, classification, database queries, live customer support, browser use, or helping Opus 5.5 and Sonnet 5.5 as a coding subagent. Anthropic says it costs about 75% less to run than Haiku 4.5 on average. Per million tokens, prompts up to 100K are $0.10 input and $0.50 output; over 100K, $0.50 and $2.50. It is the first Haiku with an adjustable effort setting. The same day, Sonnet 5.5 cache reads were halved and Max and Team subscriptions got a monthly API credit. Platform model id claude-haiku-5-5, available on the major clouds.

厂商公告Vendor announcement 2026年10月7日 Oct 7, 2026 1 官方 · 1 社区 1 official · 1 community
OpenAI OpenAI 官方 Official 社区 showcase Community showcase + 精选 + Featured

GPT-6 进 ChatGPT 聊天：回答会直接长出可点的界面（Intelligent UI） GPT-6 comes to ChatGPT chat: answers can arrive as interactive interfaces (Intelligent UI)

GPT-6 comes to ChatGPT chat: answers can arrive as interactive interfaces (Intelligent UI)

OpenAI 2026-10-07 发布：GPT-6 进入 ChatGPT 的聊天页，免费用户也有。最大的变化叫 Intelligent UI：回答不再只是文字，模型会自己决定要不要配图表、按钮、表单、地图或可交互的示意图，比如拆解一辆 7 速自行车的结构可以点着看，也能当场做一个存钱计算器或分账小工具。官方说背后是一套能边生成边显示的原生组件库加一个编译器，并专门训练了模型做版式和交互的取舍，什么时候纯文字就够也算在内。付费档（Plus、Pro、Business、Enterprise）用 GPT-6 Sol，免费和 Go 档用 GPT-6 Luna；10-07 起付费档上线，次日扩到免费档。官方还说 GPT-6 能边想边答，联网问题的 Instant 版平均早 44% 开始出字。Work 和 Codex 里的模型这次不变。 OpenAI, 2026-10-07: GPT-6 comes to ChatGPT's chat tab, free users included. The headline change is Intelligent UI: instead of text only, the model decides whether to add charts, buttons, forms, maps, or interactive diagrams, so a breakdown of a 7-speed bike becomes something you tap through, and it can build a savings calculator or bill splitter on the spot. OpenAI says this runs on a library of native, streamable components plus a compiler, and that the model was trained to make layout and interaction choices, including when plain text is enough. Paid tiers (Plus, Pro, Business, Enterprise) get GPT-6 Sol; Free and Go get GPT-6 Luna. Paid rollout starts 10-07 and expands to Free the next day. OpenAI also says GPT-6 can start answering while it keeps thinking, with Instant starting 44% sooner on average for web-search questions. Models in Work and Codex are unchanged.

厂商公告Vendor announcement 2026年10月7日 Oct 7, 2026 1 官方 · 1 社区 1 official · 1 community
Google Google 官方 Official 社区 showcase Community showcase + 精选 + Featured

Gemini 4 Argon（Google） Gemini 4 Argon (Google)

Gemini 4 Argon (Google)

Google 于 2026-09-30 宣布的下一代前沿模型 Gemini 4 Argon。官方称面向长程复杂工作流：真实软件工程、法律/金融等企业知识工作，以及网络安全防御；输出 token 上限扩到 1M（相对此前 64K）。首发只对 Fairwind Program 受信网络防御方开放；公共页写明后续尽快扩到开发者、企业与消费者，并从付费 API 与 Google AI Ultra 订户开始。入门价官方写 $2 / 百万输入 token、$10 / 百万输出 token，缓存输入相对输入价 95% off；入门期后升至 $4 / $20。官方评测摘录（标官方称，本库未复跑）：DeepSWE v1.1 77.9%；Vals Index 领先；AutomationBench 51.3%（#1）；LVBench 91.7%；CWE-bench v1 68% 并列第一。第三方榜 Artificial Analysis https://artificialanalysis.ai/models/gemini-4-argon（Gemini 4 Argon (High)，晚间 GET 200）：Intelligence Index 53（#8/223），价 $2/$10、Cache Discount 95%、每 Intelligence Index 任务约 $1.99，1M 上下文，发布日 2026-09-30；速度栏 N/A。晨间 changelog 跳过非 A/B/C 头条，未收本模型；晚间补进 Models。 Google’s next frontier model, announced 2026-09-30. Vendor claim: built for long-horizon complex workflows across real-world software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defense; output token limit raised to 1M (from the previous 64K). Initial access is limited to trusted cyber defenders in the Fairwind Program; the public page says broader access for developers, enterprises, and consumers will follow as soon as possible, starting with paid API customers and Google AI Ultra subscribers. Introductory price is stated as $2 per million input tokens and $10 per million output tokens, with cached input at 95% off the input price; after the introductory period the page says $4 / $20. Vendor eval excerpts (marked as vendor claims, not re-run here): DeepSWE v1.1 77.9%; leading on the Vals Index; AutomationBench 51.3% (#1); LVBench 91.7%; CWE-bench v1 68% tied for first. Third-party board Artificial Analysis https://artificialanalysis.ai/models/gemini-4-argon (Gemini 4 Argon (High), evening GET 200): Intelligence Index 53 (#8/223), $2/$10, Cache Discount 95%, about $1.99 per Intelligence Index task, 1M context, released 2026-09-30; speed listed N/A. Morning changelog skipped non-A/B/C headlines and did not file this model; this evening wave adds it to Models.

厂商公告Vendor announcement 2026年9月30日 Sep 30, 2026 1 官方 · 3 社区 1 official · 3 community
OpenAI OpenAI 官方 Official 社区 showcase Community showcase + 精选 + Featured

GPT-6.1 Sol（OpenAI） GPT-6.1 Sol (OpenAI)

GPT-6.1 Sol (OpenAI)

OpenAI 于 2026-09-29 发布的 GPT-6 Sol 升级款。官方称：在 agentic coding、电脑操作与专业工作上几乎对齐 GPT-6 Astra，标准输入与输出 token 价格约为 Astra 的五分之一。标准 API 价：输入 $2 / 缓存输入 $0.10 / 输出 $10 每百万 token；缓存输入官方称比标准输入少 95%、比 GPT-6 Sol 的缓存输入少 50%。API id：`gpt-6.1-sol`。可用性：Plus / Pro / Business / Enterprise / Edu 的 ChatGPT Work 与 Codex；尚未进入 Chat；API 已可用。Ultrafast 官方称将在随后数日推出，在 Codex 里相对标准速度最高 8 倍 token 生成。第三方榜 Artificial Analysis https://artificialanalysis.ai/models/gpt-6-1-sol（GPT-6.1 Sol (max)，晚间 GET 200）：Intelligence Index 52（#10/222），67.8 tok/s（#107/222），价 $2/$10、Cache Discount 95%、每 Intelligence Index 任务 $0.72，1M 上下文，发布日 2026-09-29。AA 文章称 GPT-6 Sol 在 7 天后被 6.1 Sol 接替，约比 Astra II 低 1 分、每任务成本不到其四分之一。晨间 changelog 把通用 DevDay 新闻排除在 A/B/C 之外、未收本模型（板上只有 gpt-6-astra），晚间补进 Models。 OpenAI’s upgrade to GPT-6 Sol, announced 2026-09-29. Vendor claim: it nearly matches GPT-6 Astra on agentic coding, computer use, and professional work at about one-fifth of Astra’s standard input and output token prices. Standard API price: $2 input / $0.10 cached input / $10 output per million tokens; cached input is claimed 95% less than standard input and 50% less than GPT-6 Sol’s cached input. API id: gpt-6.1-sol. Availability: Plus / Pro / Business / Enterprise / Edu in ChatGPT Work and Codex; not yet in Chat; API is available. Ultrafast is claimed for the coming days, up to 8× faster token generation versus standard speed in Codex. Third-party board Artificial Analysis https://artificialanalysis.ai/models/gpt-6-1-sol (GPT-6.1 Sol (max), evening GET 200): Intelligence Index 52 (#10/222), 67.8 tok/s (#107/222), $2/$10, Cache Discount 95%, $0.72 per Intelligence Index task, 1M context, released 2026-09-29. The AA article says GPT-6 Sol was replaced by 6.1 Sol after 7 days, about 1 point below Astra II at under a quarter of the cost per task. Morning changelog kept generic DevDay news out of A/B/C and did not file this model (the board only had gpt-6-astra); this evening wave adds it to Models.

厂商公告Vendor announcement 2026年9月29日 Sep 29, 2026 2 官方 · 3 社区 2 official · 3 community
PostHog PostHog 官方 Official 社区 showcase Community showcase + 精选 + Featured

Jeeves：带推理的开源 Jev 风格决策模型（PostHog） Jeeves: open Jev-style decision model with reasoning (PostHog)

Jeeves: open Jev-style decision model with reasoning (PostHog)

HN https://github.com/PostHog/jeeves — MIT：PostHog 开源的「会先想再决定」的 Jev 风格分类器。README：基于 Qwen3.5-9B（LoRA + pointer head），SFT + CISPO 训练，带 block-4 diffusion drafter；同一请求支持 noul / choice / score，API 与 Jev 同形（`/v1/systemone`）。作者表（thinking、greedy、2560-token 上限；Kev-9B / Jev 列为 Kev 公布数字）：Test overall 0.889 vs Kev 0.822 / Jev 0.857；JevBench overall（231 public）0.935 vs Jev 0.866；JevBench hard 0.865 vs Jev 0.730。延迟自述：无 thinking 约 0.3 s/请求，有 thinking 中位约 3.3 s（单 H100）；需 CUDA（Hopper 才用 FP8 kernel）。权重 https://huggingface.co/PostHog/jeeves。README 写 Inspired by Kev。公开仓与 HF 晚间 GET 200。本库未复跑基准、未代下权重。 README: Qwen3.5-9B with LoRA + pointer head, trained with SFT + CISPO, plus a block-4 diffusion drafter; same request supports noul / choice / score via a Jev-shaped /v1/systemone API. Author table (thinking, greedy, 2560-token cap; Kev-9B / Jev columns are Kev-published): Test overall 0.889 vs Kev 0.822 / Jev 0.857; JevBench overall (231 public) 0.935 vs Jev 0.866; JevBench hard 0.865 vs Jev 0.730. Latency claim: about 0.3 s/request without thinking, ~3.3 s median with it on one H100; needs CUDA (Hopper for the FP8 kernel). Weights at https://huggingface.co/PostHog/jeeves. README says Inspired by Kev. Public repo and HF evening GET 200. Benchmarks not re-run here; weights not downloaded for you.

厂商公告Vendor announcement 2026年9月29日 Sep 29, 2026 2 官方 · 1 社区 2 official · 1 community
Anthropic Anthropic 官方 Official 社区 showcase Community showcase + 精选 + Featured

Claude Sonnet 5.5（Anthropic） Claude Sonnet 5.5 (Anthropic)

Claude Sonnet 5.5 (Anthropic)

Anthropic 于 2026-09-28 发布的 Claude 5.5 家族第二款模型。官方页正文称：相对 Claude Sonnet 5 为明确升级，生成快 30%+，多数工作花费可低至 30%；价目与 Sonnet 5 同为输入 $2 / 输出 $10 / cache reads $0.20 每百万 token，但因 token 更少所以更便宜。定位为 Opus 5.5 的更快、更低成本补位：擅长边界清晰的日常任务、修 bug、做文档 / 幻灯 / 表格，并写有较锐的设计眼。官方表（列 Sonnet 5.5 / Sonnet 5 / Opus 5.5 / GPT-6 Sol）：Terminal-Bench 4.0 70.6% / 10.3% / 66.4%¹ / —；CursorBench 4.0 55.5% / 34.1% / 57.8% / —；GDPval-AA v2.1 1844 / 1449 / 1846 / 1487⁴。平台模型 id：`claude-sonnet-5-5`；称已在全平台可用（含 AWS / Google Cloud / Azure），零数据保留与 Opus 5.5 / Sonnet 5 相同。第三方榜 Artificial Analysis https://artificialanalysis.ai/models/claude-sonnet-5-5（变体 Adaptive Reasoning, Max Effort, Default Fallback，晚间 GET 200）：Intelligence Index 56，输出速度 138.7 tok/s（#29/216），价 $2/$10、Cache Discount 90%、每任务 $7.60，1M 上下文。晨间目录未收（当日只收 Jeff + HTML），晚间补上。 Anthropic’s second model in the Claude 5.5 family, announced 2026-09-28. The page calls it a clear upgrade over Claude Sonnet 5: 30%+ faster generation and up to 30% less cost for most work, at the same list price as Sonnet 5 — $2 input / $10 output / $0.20 cache reads per 1M tokens — because it typically uses fewer tokens. Positioned as a faster, lower-cost complement to Opus 5.5 for well-scoped everyday tasks, bug fixes, and polished documents, slides, and spreadsheets, with a noted eye for design. Vendor table (columns Sonnet 5.5 / Sonnet 5 / Opus 5.5 / GPT-6 Sol): Terminal-Bench 4.0 70.6% / 10.3% / 66.4%¹ / —; CursorBench 4.0 55.5% / 34.1% / 57.8% / —; GDPval-AA v2.1 1844 / 1449 / 1846 / 1487⁴. Platform model id: claude-sonnet-5-5; the page says it is available on all platforms including AWS / Google Cloud / Azure, with the same zero data retention as Opus 5.5 / Sonnet 5. Third-party board Artificial Analysis https://artificialanalysis.ai/models/claude-sonnet-5-5 (variant Adaptive Reasoning, Max Effort, Default Fallback, evening GET 200): Intelligence Index 56, 138.7 tok/s (#29/216), $2/$10, Cache Discount 90%, $7.60 per task, 1M context. Morning catalog skipped it (Jeff + HTML only); this evening wave picks it up.

厂商公告Vendor announcement 2026年9月28日 Sep 28, 2026 1 官方 · 2 社区 1 official · 2 community
TypeSafe AI TypeSafe AI 官方 Official 社区 showcase Community showcase + 精选 + Featured

Jev 玩宝可梦红（社区 live demo） Jev Plays Pokémon Red (community live demo)

Jev Plays Pokémon Red (community live demo)

Show HN https://jev-pokemon.vercel.app/ — 社区 live：用 TypeSafe Jev 决策模型通关《宝可梦 红》。作者称 harness 只读 Game Boy RAM、列出合法选项与事实，由 Jev 选一项再按键；无脚本通关、不写内存。直播页自称默认静音、右侧显示每次决策与概率；页内有 Frigade 推广条，且写明与任天堂 / Game Freak / The Pokémon Company 无关、不附带 ROM。本卡只记 Jev 能力展示，不是可下载的游戏 ROM。 The author says the harness only reads Game Boy RAM, lists legal options with facts, and lets Jev pick one before pressing buttons; no scripted win path, no memory writes. The live page says it starts muted and shows each decision and odds on the right; it also carries a Frigade promo bar and states it is not affiliated with Nintendo / Game Freak / The Pokémon Company and ships no ROM. Catalogued as a Jev capability showcase only, not a downloadable game ROM.

厂商公告Vendor announcement 2026年9月25日 Sep 25, 2026 1 官方 · 3 社区 1 official · 3 community
ollaya-dev ollaya-dev 官方 Official 社区 showcase Community showcase + 精选 + Featured

Ollaya：本机跑开源决策模型（Jev / Laya 相邻） Ollaya: run open decision models locally (Jev / Laya-adjacent)

Ollaya: run open decision models locally (Jev / Laya-adjacent)

HN 首页 https://ollaya.dev/ — 作者定位「像 Ollama 跑 LLM 一样，本机跑开源决策模型」。拉 Laya / decider / NLI / GLiClass 等开源权重，本地 daemon 提供 TypeSafe 兼容的 /v1/systemone；官方称可改 TYPESAFE_BASE_URL 指向本机，现有 Jev 客户端不用改代码。延迟对比数字（如 Laya 8–10ms、Jev hosted 236–276ms）一律标官网自述 / 第三方基准，本库未复跑。 the site says existing Jev clients work by pointing TYPESAFE_BASE_URL at localhost. Latency comparisons (e.g. Laya 8–10ms, Jev hosted 236–276ms) are site or third-party claims — not re-run here. not a separate card.

厂商公告Vendor announcement 2026年9月25日 Sep 25, 2026 2 官方 · 1 社区 2 official · 1 community
Anthropic Anthropic 官方 Official 社区 showcase Community showcase + 精选 + Featured

Claude Opus 5.5（Anthropic） Claude Opus 5.5 (Anthropic)

Claude Opus 5.5 (Anthropic)

Anthropic 于 2026-09-22 发布的 Claude 5.5 家族第一款模型。官方页正文称：它在多数工作上达到 Claude Fable 5.1 的水平，按典型负载跑起来比 Opus 5 少 40% 花费；meta description 称它在 agentic coding 与知识工作上领先，且比 Opus 5 少 40% 运行成本。官方价目（每百万 token）：输入 $4、输出 $20（称比 Opus 5 少 20%），cache reads $0.20（对照 Opus 5 的 $0.50，称少 60%），cache writes $5（对照 $6.25）；fast mode 为输入 $8 / 输出 $40，称最高 2.5 倍速度。第三方榜：Artificial Analysis 页面（变体 “Adaptive Reasoning, Max Effort, Default Fallback”）列出 Intelligence Index 58（#1 / 211）、输入 $4.00 / 输出 $20.00、缓存折扣 95%、每任务 $5.98、1M 上下文；速度栏为 N/A。 Anthropic’s first model in the Claude 5.5 family, announced 2026-09-22. The page says it performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5 on typical workloads; the meta description says it leads in agentic coding and knowledge work and costs 40% less to run than Opus 5. Vendor price table (per 1M tokens): input $4, output $20 (claimed 20% less than Opus 5), cache reads $0.20 (vs Opus 5 at $0.50, claimed 60% less), cache writes $5 (vs $6.25); fast mode is $8 input / $40 output, claimed up to 2.5× speed. Third-party board: the Artificial Analysis page (variant “Adaptive Reasoning, Max Effort, Default Fallback”) lists Intelligence Index 58 (#1 / 211), $4.00 input / $20.00 output, a 95% cache discount, $5.98 per task, and a 1M context; the speed cell is N/A.

厂商公告Vendor announcement 2026年9月22日 Sep 22, 2026 1 官方 · 2 社区 1 official · 2 community
Xiaomi MiMo Xiaomi MiMo 官方 Official 社区 showcase Community showcase + 精选 + Featured

MiMo V2.6（小米 MiMo） MiMo V2.6 (Xiaomi MiMo)

MiMo V2.6 (Xiaomi MiMo)

小米 MiMo 于 2026-09 发布的 MiMo-V2.6 系列旗舰（官方页 meta：「Introducing the MiMo-V2.6 series: frontier intelligence, all the modalities, built in public.」）。Hugging Face 卡 XiaomiMiMo/MiMo-V2.6-Pro-RL 官方称：系列旗舰，主题「Scaling Reinforcement Learning Toward Self-Improvement」；稀疏 MoE，1.02T 总参数 / 42B 激活；原生全模态（文本、图像、视频、音频输入），1M 上下文。同系列 HF 还有 MiMo-V2.6-Flash-RL、MiMo-V2.6-Distill-Qwen-9B。第三方榜：Artificial Analysis 页面列出 MiMo-V2.6-Pro Intelligence Index 46、开源权重、MIT、1M 上下文、多模态输入（AA 榜单数据，非本库复跑）。 Flagship of Xiaomi MiMo’s MiMo-V2.6 series (released September 2026; official page meta: “Introducing the MiMo-V2.6 series: frontier intelligence, all the modalities, built in public.”). The Hugging Face card XiaomiMiMo/MiMo-V2.6-Pro-RL states vendor claims: series flagship under the theme “Scaling Reinforcement Learning Toward Self-Improvement”; sparse MoE with 1.02T total / 42B activated parameters; native omnimodal inputs (text, image, video, audio) and a 1M-token context. Series siblings on HF include MiMo-V2.6-Flash-RL and MiMo-V2.6-Distill-Qwen-9B. Third-party board: the Artificial Analysis page lists MiMo-V2.6-Pro Intelligence Index 46, open weights, MIT, 1M context, and multimodal inputs (AA’s figures, not re-run here). optionally 49796660 (AA analysis).

厂商公告Vendor announcement 2026年9月22日 Sep 22, 2026 2 官方 · 3 社区 2 official · 3 community
StepFun StepFun 官方 Official 社区 showcase Community showcase + 精选 + Featured

Step 5 Preview（阶跃星辰 StepFun） Step 5 Preview (StepFun)

Step 5 Preview (StepFun)

阶跃星辰 2026-09-20 发布的新旗舰，官方定位「面向 agentic 工作」：软件工程、Agent 任务、专业知识工作，尤其强调金融。官方称稀疏 MoE、600B 总参数 / 每 token 27B 活跃、1M 上下文、原生图像输入；当日经产品与 API 开放，开源权重定档 2026-10-15。第三方榜：Artificial Analysis Intelligence Index 44（AA 榜单数据，非本库复跑）。注意：Hugging Face 上的 stepfun-ai/Step-5-Preview-BF16 今晚返回 401 / 仅占位，权重尚不可下载。 StepFun’s new flagship, announced 2026-09-20 and pitched by the vendor as a model “for agentic work”: software engineering, agentic tasks, professional knowledge work, with particular emphasis on finance. Vendor claims: sparse MoE with 600B total / 27B active parameters per token, a 1M-token context window, and native image input; available the same day via StepFun’s products and API, with open weights scheduled for 2026-10-15. Third-party board: Artificial Analysis Intelligence Index 44 (AA’s figure, not re-run here). Note: the Hugging Face repo stepfun-ai/Step-5-Preview-BF16 returned 401 / a placeholder tonight — weights are not downloadable yet.

厂商公告Vendor announcement 2026年9月20日 Sep 20, 2026 1 官方 · 3 社区 1 official · 3 community
TypeSafe AI TypeSafe AI 官方 Official 社区 showcase Community showcase + 精选 + Featured

Jev / System One Models（TypeSafe AI） Jev / System One Models (TypeSafe AI)

Jev / System One Models (TypeSafe AI)

TypeSafe AI 于 2026-09-15 发布的首个公开 System One Model。面向软件内决策而非聊天：输入非结构化状态，输出类型安全的结构化决策，并附带校准概率与置信度。官方称新架构 + 并行采样 + Reinforcement Learning for Calibrated Decisions（RLCD）；目前为早期访问；速度、价格与评测倍率均为厂商自述，本库未复跑其工作流评测。 TypeSafe AI’s first public System One Model, announced 2026-09-15. Built for in-software decisions rather than chat: unstructured state in, type-safe structured decisions out with calibrated probabilities and confidence. Vendor claims a new architecture, parallel sampling, and Reinforcement Learning for Calibrated Decisions (RLCD); currently early access; speed, pricing, and eval multipliers are all vendor claims, and the workflow evals were not re-run by this library.

厂商公告Vendor announcement 2026年9月15日 Sep 15, 2026 2 官方 · 1 社区 2 official · 1 community
HAL-X AI HAL-X AI 官方 Official 社区 showcase Community showcase

THX-01：322M 开源决策模型，一次前向给出带校准概率的选项 THX-01: a 322M open decision model that returns calibrated choices in one forward pass

THX-01: a 322M open decision model that returns calibrated choices in one forward pass

HAL-X AI（阿塞拜疆）在 2026-10-10 开源的 THX-01：约 3.22 亿参数的非自回归决策模型，不写长文，只对你给出的现状和若干类型化问题（选择题、是否、打分、抽数字、摘原文、找依据）一次前向返回带校准概率的结构化答案。权重与 Python 包 `thx01` 为 Apache-2.0；另有托管 API 与 Hugging Face Space 演示。作者在 HF 博文里用内部多语工单基准对比 TypeSafe Jev，并给出延迟等数字——本库没有复跑。适合路由、分流、抽取这类「只要一个决定」的 Agent 环节。 THX-01 from HAL-X AI (Azerbaijan), open-sourced 2026-10-10: a ~322M non-autoregressive decision model that does not write long prose. You pass a state plus typed questions (choice, yes/no, score, number extract, excerpt, citation) and get structured answers with calibrated probabilities in one forward pass. Weights and the Python package `thx01` are Apache-2.0; there is also a hosted API and a Hugging Face Space demo. The HF blog compares it to TypeSafe Jev on an internal multilingual ticket benchmark with latency figures—this library did not re-run them. Good for agent steps that only need a decision: routing, triage, extraction.

厂商公告Vendor announcement 2026年10月10日 Oct 10, 2026 3 官方 · 1 社区 3 official · 1 community
Google DeepMind Google DeepMind 官方 Official 社区 showcase Community showcase

EmbeddingGemma 2：把文字、图片、声音和视频收成同一套向量的小模型 EmbeddingGemma 2: a small model that turns text, images, audio, and video into one vector space

EmbeddingGemma 2: a small model that turns text, images, audio, and video into one vector space

EmbeddingGemma 2 是 Google DeepMind 在 2026-10-06 发布的开源嵌入模型。嵌入是一串数字，用来表示一段文字、一张图或一段声音的意思，方便按意思搜索，而不是按关键词硬对。这一版把文字（含代码）、图片、视频和声音放进同一个 768 维的空间。模型卡写总参数 7.4 亿，其中文字部分约 2.7 亿，视觉编码器 1.7 亿，音频编码器 3 亿，可以只加载用得到的那一部分。许可页标题是 Apache License 2.0。它面向手机和笔记本上的本地搜索。模型卡里的测试分是 Google 自己报的，本库没有复跑，也没有下载权重。 EmbeddingGemma 2 is an open embedding model Google DeepMind released on 2026-10-06. An embedding is a list of numbers that stands for the meaning of a passage, a picture, or a sound, so you can search by meaning rather than by exact keywords. This version puts text (including code), images, video, and audio in one 768-dimension space. The model card says 740 million parameters in total: about 270 million for text, a 170 million vision encoder, and a 300 million audio encoder, and you can load only the part you need. The license page is titled Apache License 2.0. It is aimed at search on a phone or laptop. Scores on the model card are Google’s own. This library did not re-run them and did not download the weights.

厂商公告Vendor announcement 2026年10月6日 Oct 6, 2026 4 官方 · 1 社区 4 official · 1 community
Google Google 官方 Official 社区 showcase Community showcase

Nano Banana 2.1：Google 新版出图改图模型，主打海报排版、局部改图和角色一致 Nano Banana 2.1: Google’s new image generation and editing model, aimed at layout design, masked edits, and consistent subjects

Nano Banana 2.1: Google’s new image generation and editing model, aimed at layout design, masked edits, and consistent subjects

Nano Banana 2.1 是 Google 在 2026 年 10 月 6 日（北京时间 10 月 7 日凌晨）发布的出图和改图模型，是 Nano Banana 2 的升级版。适合要做海报、信息图、商品图、角色设定，并且需要反复修改的设计师和内容创作者。Google 在 X 上说它全面超过自家之前的图片模型，进步最明显的是三件事：版式设计、用遮罩改图（先圈出一块区域，只改这一块），以及同一个人物或商品在多张图里保持一致。普通用户可以在 Gemini App 里试（官方模型卡把它列为发布渠道，正在陆续上线）；开发者可以在 Google AI Studio 里试，或用 API 调用 gemini-nano-banana-2.1，API 没有免费档。下面的对比分数和价格都是 Google 自己公布的，本库没有实际出图、没有复跑对比。 Nano Banana 2.1 is Google’s image generation and editing model, released on 6 October 2026 (early 7 October Beijing time) as an upgrade to Nano Banana 2. It suits designers and content creators who make posters, infographics, product shots, or character sheets and need to iterate. Google’s X post says it beats the company’s previous image models across the board, with the biggest gains in visual design, mask-based editing (outline an area and change only that part), and keeping the same person or product consistent across images. Regular users can try it in the Gemini app (listed as a launch channel on the model card, rolling out); developers can try it in Google AI Studio or call gemini-nano-banana-2.1 through the API, which has no free tier. The comparison scores and prices below are Google’s own; this library did not generate images or re-run any comparison.

厂商公告Vendor announcement 2026年10月6日 Oct 6, 2026 7 官方 · 1 社区 7 official · 1 community
Reflection Reflection 官方 Official 社区 showcase Community showcase

Beam：Reflection 的 501B 开源权重模型（还不能下载） Beam: Reflection’s 501B open-weight model (not downloadable yet)

Beam: Reflection’s 501B open-weight model (not downloadable yet)

Beam 是 Reflection 在 2026 年 10 月 5 日公开的开源权重模型，给要写代码、做推理、让编程助手干活的人用。它是稀疏混合专家：参数一共 5010 亿，每次真正计算时只用其中 230 亿。现在还不能下载。官网写它还在做最后一轮安全测试（专门找它会不会被带去干不该干的事），眼下只能到平台页登记提前试用；权重、技术报告、模型卡和给开发者的配套，都说这个月晚些时候才放，并打算用 Apache 2.0。页面表格里的分数是厂商自己报的，这里没有复跑。 Beam is Reflection’s open-weight model, announced on 5 October 2026, for people who write code, reason through problems, or hand work to a coding agent. It is a sparse mixture-of-experts model: 501 billion parameters in total, with 23 billion active on each step. It is not downloadable yet. The site says it is still in final safety testing (checking whether it can be talked into doing something it should not), and that early access is a signup only. Weights, the technical report, the model card, and the developer extras are promised later this month, under Apache 2.0. The scores on the page are the vendor’s own numbers and were not re-run here.

厂商公告Vendor announcement 2026年10月5日 Oct 5, 2026 2 官方 · 1 社区 2 official · 1 community
Cloudflare Cloudflare 官方 Official 社区 showcase Community showcase

Clef：Cloudflare 开源决策模型（含 Clef-Flash） Clef: Cloudflare open-weight decision models (incl. Clef-Flash)

Clef: Cloudflare open-weight decision models (incl. Clef-Flash)

Cloudflare 发布开源决策模型 Clef，以及更小的 Clef-Flash。它们不生成自由文本：读入一段状态和一组带类型的问题，为每个允许的答案返回概率。Workers AI 上的模型 id 是 @cf/cloudflare/clef（27B）和 @cf/cloudflare/clef-flash（9B）。官方称接口与 Jev / System One 同形，问题类型仍是 noul / choice / score，并带视觉编码器，状态可以是文本、JSON、图像或视频。权重在 Hugging Face，许可 Apache-2.0。HF 卡写 Clef 由 Qwen/Qwen3.8-27B 后训练、Clef-Flash 由 Qwen/Qwen3.5-9B 后训练，这是厂商 / HF 口径。官博还宣布 RL 微调：先由 forward-deployed engineer 动手，之后再做成自助平台——按路线图记，不是已经核过的产品深度。Decision Index、延迟和工作流分数一律是厂商自述，本库未复跑。同板与 Jev、Laya、Jeeves、Ollaya 相邻；Cases 里已有本机 Jeff 和命令行 Onesie。 Cloudflare released the open-weight decision model Clef and the smaller Clef-Flash. They do not generate free text: they read a state plus typed questions and return a probability for every allowed answer. On Workers AI the ids are @cf/cloudflare/clef (27B) and @cf/cloudflare/clef-flash (9B). The vendor says the API matches Jev / System One, with noul / choice / score questions and a vision encoder, so state can be text, JSON, images, or video. Weights are on Hugging Face under Apache-2.0. The HF cards say Clef is post-trained from Qwen/Qwen3.8-27B and Clef-Flash from Qwen/Qwen3.5-9B — a vendor / HF claim. The blog also announces RL fine-tuning: hands-on with a forward-deployed engineer first, then a later self-serve platform. That is a roadmap, not a product depth this library verified. Decision Index, latency, and workflow scores are vendor claims and were not re-run here. On this board it sits next to Jev, Laya, Jeeves, and Ollaya; Cases already lists local Jeff and the Onesie CLI.

厂商公告Vendor announcement 2026年10月1日 Oct 1, 2026 4 官方 · 1 社区 4 official · 1 community
kouhxp kouhxp 官方 Official 社区 showcase Community showcase

Gutsy：本机 CPU 可跑的 0.8B Jev 兼容决策模型 Gutsy: 0.8B Jev-compatible decision model for local CPU

Gutsy: 0.8B Jev-compatible decision model for local CPU

kouhxp 开源的 gutsy-0.8b（v0.4）：把 Qwen3.5-0.8B 微调成决策模型，不生成自由文本。你送一段状态（文本或 JSON）和 typed questions（yes/no、choice、score），它为每个选项返回校准概率；choice / score 还带 reject。运行时 gutsy-inference 走 llama.cpp，默认可在普通 CPU 上起本地 HTTP；接口刻意对齐 Jev / System One（`/v1/systemone` 等），现有客户端多半只换 base URL。权重在 Hugging Face kouhxp/gutsy（Q8_0 与 Q4_K_M GGUF），代码与权重 README 写 Apache-2.0。JevBench / 评测表与延迟是作者自测，本库未复跑。同板与 Jev、Laya、Clef、Jeeves、Ollaya 相邻；Cases 已有本机 Jeff 与 Onesie CLI。 kouhxp’s open gutsy-0.8b (v0.4) fine-tunes Qwen3.5-0.8B into a decision model that does not generate free text. You send a state (text or JSON) plus typed questions (yes/no, choice, score) and get a calibrated probability for every option; choice / score answers also carry reject. The gutsy-inference runtime uses llama.cpp and can serve a local HTTP API on an ordinary CPU. The API deliberately matches Jev / System One (`/v1/systemone` and related), so many existing clients only need a new base URL. Weights live on Hugging Face kouhxp/gutsy (Q8_0 and Q4_K_M GGUF); README says Apache-2.0 for code and weights. JevBench / eval tables and latency are author self-tests, not re-run here. On this board it sits next to Jev, Laya, Clef, Jeeves, and Ollaya; Cases already lists local Jeff and the Onesie CLI.

厂商公告Vendor announcement 2026年10月1日 Oct 1, 2026 2 官方 · 1 社区 2 official · 1 community
Strands Agents Strands Agents 官方 Official 社区 showcase Community showcase

Strands Decider 2B：一个小的开源决策模型，只做选择题和打分 Strands Decider 2B: a small open decision model that picks an option or gives a score

Strands Decider 2B: a small open decision model that picks an option or gives a score

Strands Decider 2B 是 Strands Agents 放出的一个小决策模型。它不写文章。你给它一段现状和几个允许的答案，它为每个答案打一个概率，也可以按一个分数尺打分。仓库写它大约 19 亿参数，基于 Qwen3.5-2B 改来，拿掉了原来用来生成文字的那一层，换上一个大约一百万参数的小头。代码和权重以 Apache-2.0 发布，可以在自己的 CPU 或显卡上跑。官博是 2026-10-01，仓库现在指向的参考版本是 v21，更早的 v19 仍在。本库没有安装，也没有复跑仓库里的分数。 Strands Decider 2B is a small decision model from Strands Agents. It does not write prose. You give it a situation and the answers you allow, and it returns a probability for each one, or a score on a scale. The repo says it has about 1.9 billion parameters, starts from Qwen3.5-2B, removes the layer that used to generate text, and adds a small head of about a million parameters. Code and weights are Apache-2.0, and it can run on your own CPU or GPU. The blog is dated 2026-10-01. The repo’s reference version now is v21, and the earlier v19 is still published. This library did not install it and did not re-run the repo’s scores.

厂商公告Vendor announcement 2026年10月1日 Oct 1, 2026 4 官方 · 1 社区 4 official · 1 community
OpenAI OpenAI 官方 Official 社区 showcase Community showcase

Astra 重建《疯狂大厦》3D 漫步（社区 demo） Astra rebuilds Maniac Mansion as a 3D walkthrough (community demo)

Astra rebuilds Maniac Mansion as a 3D walkthrough (community demo)

Show HN https://github.com/s-macke/maniac-mansion-3d — 作者 s-macke 在 HN 自述：先用 GPT-6 Astra + Blender 做真实物体重建实验，再把 DOS EGA《Maniac Mansion》原画屏幕喂给 Astra 转成 3D 场景，最终做成整栋宅的浏览器漫步。Live：https://s-macke.github.io/maniac-mansion-3d/。README：34 个房间空间、Three.js、烘焙光照、非可玩重制（无谜题/背包）。仓库描述写 Built with Blender, Three.js and AI。本卡只记 Astra 能力展示与可点开的 walkthrough，不为 LucasArts / Disney IP 背书，也不分发原作资源。 Live: https://s-macke.github.io/maniac-mansion-3d/. README: 34 room spaces, Three.js, baked lighting, exploration only (no puzzles/inventory). Repo blurb: Built with Blender, Three.js and AI. Catalogued as an Astra capability showcase with a clickable walkthrough — not an endorsement of LucasArts / Disney IP, and no original game assets are redistributed here.

厂商公告Vendor announcement 2026年9月27日 Sep 27, 2026 1 官方 · 3 社区 1 official · 3 community
Fireworks Fireworks 官方 Official 社区 showcase Community showcase

Ember-1（Fireworks Research 专用模型） Ember-1 (Fireworks Research specialized model)

Ember-1 (Fireworks Research specialized model)

Fireworks Research 的专用模型 Ember-1。官博 https://fireworks.ai/blog/ember-1 页眉 PUBLISHED 9/23/2026，模型页 https://fireworks.ai/models/fireworks/ember-1 元数据 Created on 9/22/2026。模型页写 model path：accounts/fireworks/models/ember-1。官方自述：基于 Kimi K3，推理更短，大约少用 40% token 且质量相当；Serverless 定价 $3.00 / $0.30 / $15.00 每百万 token（input / cached input / output）；Mixture-of-Experts、参数 2.78T、上下文 1040k；Function Calling 与 image input 为 Supported，Fine-tuning 为 Not supported；官博写 “Available today”。晨间目录未收，晚间补上。 Ember-1, a specialized model from Fireworks Research. The blog at https://fireworks.ai/blog/ember-1 is marked PUBLISHED 9/23/2026; the model page at https://fireworks.ai/models/fireworks/ember-1 lists Created on 9/22/2026. The model page lists model path accounts/fireworks/models/ember-1. Vendor claims: built on Kimi K3, with shorter reasoning, about 40% fewer tokens at comparable quality; serverless pricing $3.00 / $0.30 / $15.00 per 1M tokens (input / cached input / output); Mixture-of-Experts, 2.78T parameters, 1040k context; function calling and image input Supported, fine-tuning Not supported. The morning catalog skipped it; this evening wave picks it up.

厂商公告Vendor announcement 2026年9月23日 Sep 23, 2026 2 官方 · 1 社区 2 official · 1 community
ConvAI Innovations ConvAI Innovations 官方 Official 社区 showcase Community showcase

Laya：开源 System 1 决策引擎（对照 Jev） Laya: open-source System 1 decision engine (Jev-adjacent)

Laya: open-source System 1 decision engine (Jev-adjacent)

作者称 Laya 是非自回归、亚 35ms 的开源 System 1 决策模型族（Apache 2.0 权重），提供 choice / score / noul 三类结构化决策与多语路由；对照站内已收录的 TypeSafe Jev。基准、延迟、与 Jev 对比数字一律标「作者称 / 第三方整理」，本库未复跑。HN 讨论：https://news.ycombinator.com/item?id=49765348（2026-09-19）。 Author presents Laya as an open-weight non-autoregressive System 1 decision family (Apache 2.0) with choice/score/noul primitives and multilingual routing; sits next to catalogued TypeSafe Jev. Latency/benchmark/Jev-comparison figures are author or third-party claims — not re-run here.

厂商公告Vendor announcement 2026年9月19日 Sep 19, 2026 1 官方 · 2 社区 1 official · 2 community
OpenAI OpenAI 官方 Official 社区 showcase Community showcase

GPT-6 Astra 官方 Showcase + 社区合集 GPT-6 Astra official Showcase + community roundup

GPT-6 Astra official Showcase + community roundup

官方展示墙 + 中文入口帖。OpenAI Developers 站的 Showcase 页（developers.openai.com/showcase，2026-09-13 本库打开为 200）是官方 demo 展示墙，分 Games / Apps / Websites / API 四栏，并带按模型筛选（含 GPT-6 Astra）。宝玉（@dotey）9 月 12 日转发 OpenAIDevs 的「GPT-6 Astra by our community」社区合集文，称「GPT-6 Astra 社区作品集合，评论也有不少优秀作品」，并指向该 Showcase 页。OpenAIDevs 文内列举的作品（3D 解剖部件、Unreal Engine 复刻曼哈顿、梵高六幅画变 Three.js 可走小镇、AR 吸尘覆盖图、Blender 火车模型、浏览器赛车等）是官方社区文自述，本库未逐条打开。本卡只记「可逛的官方展示墙 + 中文入口帖」，不编造作品条数。 An official demo wall plus a Chinese entry post. The Showcase page on OpenAI Developers (developers.openai.com/showcase, returned 200 when this library opened it on 2026-09-13) is OpenAI’s own demo wall with Games / Apps / Websites / API tabs and a model filter that includes GPT-6 Astra. On 12 Sept 宝玉 (@dotey) quoted OpenAIDevs’ “GPT-6 Astra by our community” roundup article, wrote “a collection of GPT-6 Astra community builds — the replies have good ones too,” and pointed to that Showcase page. The builds listed in the OpenAIDevs article (3D anatomical parts, a Manhattan recreation in Unreal Engine, six Van Gogh paintings as a walkable Three.js town, AR vacuum-coverage mapping, a Blender train model, a browser racing game, and more) are the official community article’s own descriptions; not opened one by one here. This card records a browsable official wall plus a Chinese entry post only — no invented build counts.

厂商公告Vendor announcement 2026年9月12日 Sep 12, 2026 2 官方 · 2 社区 2 official · 2 community
DeepSeek DeepSeek 官方 Official 社区 showcase Community showcase

DeepSeek-V4.1-Flash（DeepSeek） DeepSeek-V4.1-Flash (DeepSeek)

DeepSeek-V4.1-Flash (DeepSeek)

DeepSeek 于 2026-09-10 正式发布的新架构家族最小型号，原生多模态视觉理解。官方称 552B MoE、非对称 Causal Encoder–Decoder（输入约 8B / 输出约 16B 活跃参数），API 模型名 `deepseek-flash`；旧名 `deepseek-v4-flash` / `deepseek-v4-flash-vision-exp` 暂时路由到本模型。官方称多项测试上优于 V4-Pro，计划自 2026-09-14 北京时间 12:00 起将 `deepseek-v4-pro` 请求路由到 V4.1-Flash 并按 Flash 计价，直至 V4.1-Pro。基准分与缓存压缩数字一律标「官方称」，本库未复跑。 DeepSeek’s new-architecture family’s smallest model, officially released 2026-09-10, with native multimodal visual understanding. Vendor claims: 552B-parameter MoE with an asymmetric Causal Encoder–Decoder (about 8B active parameters for input / 16B for output); API model name `deepseek-flash`; the old names `deepseek-v4-flash` / `deepseek-v4-flash-vision-exp` temporarily route here. DeepSeek says it beats V4-Pro on multiple tests and plans, from 12:00 Beijing time on 2026-09-14, to route `deepseek-v4-pro` requests to V4.1-Flash at Flash pricing until V4.1-Pro ships. Benchmark and cache-compression figures are all labelled vendor claims; not re-run by this library.

厂商公告Vendor announcement 2026年9月10日 Sep 10, 2026 3 官方 · 1 社区 3 official · 1 community
OpenAI OpenAI 官方 Official 社区 showcase Community showcase

Tripo3D 上的 Astra 3D 提示词库（社区资源） Astra 3D prompt gallery on Tripo3D (community resource)

Astra 3D prompt gallery on Tripo3D (community resource)

社区资源：@cnyzgkc（木马人）9 月 9 日发帖称「分享一个 GPT-6 Astra 3D 提示词网站，拥有 200+ 的提示词，而且完全免费」，附约 28 秒录屏，链接指向 Tripo3D 的 Astra 3D 提示词页（中文路径）。该页汇集社区用 GPT-6 Astra 做出的 Blender 场景、Three.js 网站与浏览器游戏案例，每条列作品预览、作者、来源与提示词，并以「逐字 Prompt / 基于来源整理」标注证据等级，含 GitHub 仓库的案例可筛选。「200+」是发帖人的说法（录屏封面帧显示 201 条；本库 9 月 11 日打开时页面显示 215 条），本库未逐条复数。本卡只记第三方提示词画廊，不是模型卡，也不为 Tripo3D 产品背书。 Community resource: on 9 Sept @cnyzgkc (木马人) posted “sharing a GPT-6 Astra 3D prompt site with 200+ prompts, completely free,” with a ~28s screen recording linking to Tripo3D’s Astra 3D prompts page (Chinese path). The page collects community Blender scenes, Three.js sites, and browser games built with GPT-6 Astra; each entry lists a preview, author, source, and prompt, with evidence labelled “verbatim prompt” vs “reconstructed from source,” and entries with GitHub repos can be filtered. “200+” is the poster’s claim (the recording’s poster frame shows 201; the page showed 215 when this library opened it on 11 Sept) — not re-counted item by item here. A third-party prompt-gallery card, not a model card, and no endorsement of Tripo3D’s product.

厂商公告Vendor announcement 2026年9月9日 Sep 9, 2026 1 官方 · 1 社区 1 official · 1 community
OpenAI OpenAI 官方 Official 社区 showcase Community showcase

Astra 复刻迷你 LOL（社区 demo） Astra recreates a mini League of Legends (community demo)

Astra recreates a mini League of Legends (community demo)

社区 demo：@konstiwohlwend 9 月 8 日发帖称，让 GPT-6 Astra「照抄英雄联盟」，它真做出来了，附约 44 秒视频；封面帧是选人界面，Garen / Ashe / Ahri / Lux / Lee Sin 五英雄、5v5（你 + 4 队友对 5 个机器人）。中文转述来自 @LufzzLiz（岚叔）同日帖，并把作者两步提示词译成中文：先「做一个跟 LOL 一样的游戏，先 5 个英雄」，再骂回去「这是廉价山寨，先写计划、按真实尺寸与机制实现、UI 要原生不许 HTML 贴上去」。本卡只记模型能力展示，不是可下载的游戏。 Community demo: on 8 Sept @konstiwohlwend posted that he asked GPT-6 Astra to copy League of Legends “and it actually did,” with a ~44s video; the poster frame is a champion-select screen with Garen / Ashe / Ahri / Lux / Lee Sin, 5v5 (you + 4 allies vs 5 bots). The Chinese write-up is @LufzzLiz’s same-day post, which also renders the author’s two-step prompt in Chinese: first “build a game exactly like LoL, start with 5 champions,” then push back “this is a cheap clone — plan first, implement real scale and mechanics, UI must be native, not HTML pasted on top.” Catalogued as a model-capability showcase only, not a downloadable game.

厂商公告Vendor announcement 2026年9月8日 Sep 8, 2026 1 官方 · 2 社区 1 official · 2 community
OpenAI OpenAI 官方 Official 社区 showcase Community showcase

GPT-6 Astra（OpenAI） GPT-6 Astra (OpenAI)

GPT-6 Astra (OpenAI)

OpenAI 2026-09-03 发布的前沿模型。官方定位：电脑操作、浏览、软件工程、网络安全、科研与专业办公；ChatGPT 内可用 Sites 从一条提示直接生成并托管网站、Web 应用与游戏；API 名 gpt-6-astra。社区侧：YouWare 9 月 8 日发帖称已整理 720+ 条 Astra 社区作品（游戏原型、3D 世界、产品 UI、动效、模拟），附提示词可在 YouWare 复跑。 OpenAI’s frontier model, announced 2026-09-03. Official pitch: computer use, browsing, software engineering, cybersecurity, science, and professional work; in ChatGPT, Sites builds and hosts websites, web apps, and games from one prompt; API name gpt-6-astra. Community side: on 8 Sept YouWare posted that it sorted 720+ community Astra builds (game prototypes, 3D worlds, product UIs, motion, sims), prompts included to replay on YouWare.

厂商公告Vendor announcement 2026年9月3日 Sep 3, 2026 3 官方 · 2 社区 3 official · 2 community
Anthropic Anthropic 官方 Official

Claude Fable 5.1 / Mythos 5.1（Anthropic） Claude Fable 5.1 / Mythos 5.1 (Anthropic)

Claude Fable 5.1 / Mythos 5.1 (Anthropic)

Anthropic 2026-09 发布 Claude Fable 5.1（通用）与 Mythos 5.1（受信访问）：同一底层模型、不同防护等级。定位编码、知识工作与长任务 / Agent，并展示科研侧早期例子。已在 Grok Bot 库以「AI 资讯」收录，这里只作为模型看板的一张卡，正文见库内条目。 Anthropic’s Sept 2026 launch of Claude Fable 5.1 (generally available) and Mythos 5.1 (trusted access) — same underlying model, different safeguard levels. Pitch is coding, knowledge work, and long-running / agentic tasks, plus early research demos. Already catalogued as AI news in the Grok Bot library; this is the board card, full copy lives on that entry.

厂商公告Vendor announcement 2026年9月1日 Sep 1, 2026 2 官方 · 0 社区 2 official · 0 community