#kimik25 — Public Fediverse posts

https://winbuzzer.com/2026/05/18/cursor-releases-composer-25-saying-its-better-at-s-xcxwbn/

Cursor has launched Composer 2.5: The upgrade improves long-running coding work and instruction following without switching away from the Kimi K2.5 base used for Composer 2.

#AI #Composer2 #Cursor #KimiK25 #Coding #AICoding #AIAgents #AgenticAI #AIModels #CodeGeneration #DeveloperTools #Developers

https://winbuzzer.com/2026/05/18/cursor-releases-composer-25-saying-its-better-at-s-xcxwbn/

Cursor has launched Composer 2.5: The upgrade improves long-running coding work and instruction following without switching away from the Kimi K2.5 base used for Composer 2.

#AI #Composer2 #Cursor #KimiK25 #Coding #AICoding #AIAgents #AgenticAI #AIModels #CodeGeneration #DeveloperTools #Developers

Как использовать Claude Code в 8 раз дешевле: подключаем китайские модели

Всем привет! Сегодня разберём, можно ли использовать Claude Code с китайскими моделями вместо Opus и сколько на этом реально сэкономить. Взял две китайские модели Kimi K2.6 от Moonshot AI и GLM-5.1 от Z AI и прогнал их на привычных задачах и сравнил с Opus. Не в написании кода, а на повседневных задачах: создать лендинг, сделать карусель для соцсетей, анализ данных, что-то в интернете поискать сравнить, написать какой-нибудь не сложный Telegram-бот и т.д. Забегая вперёд: результаты оказались неожиданными. Где-то китайские модели выиграли у Opus, где-то проиграли, а где-то разница оказалась чисто вкусовой. В конце статьи будет пошаговая инструкция, как подключить любую из этих моделей к Claude Code за пять минут. Работает и в терминале, и в десктоп-приложении, и в VS Code, везде где вы привыкли запускать Claude Code

https://habr.com/ru/articles/1026760/

#claude #claude_code #kimik2 #kimik25 #openai #ai #ии_агенты

Ein API-Leak entlarvt Kimi K2.5 von Moonshot AI als technische Open-Source-Basis des neuen Cursor Composer 2.

Cursor nutzte das Pretraining, optimierte die Architektur aber durch eigenes Fine-Tuning und Reinforcement Learning für extrem lange Code-Eingaben und ein effizienteres Token-Handling.

#Cursor #OpenSource #KimiK25 #KIModelle #News
https://www.all-ai.de/news/news26/cursor-composer-kimi

Из коробки не работает: запускаем свежие большие LLM

В последнее время открытых моделей сверхбольшого размера развелось неимоверное количество, даже не просто моделей, а производителей. Вариации GLM, Kimi, DeepSeek занимают по нескольку строк в топ 5-10-20. Понадобилось перебрать основные LLM для тестов и выбора "рабочей лошадки", для чего пришлось немного пошуршать в интернетах. Оставлю в качестве памятки, вдруг кому-то окажется полезным. Всё делалось на базе образов vllm-openai, платформ B200/H200 и дров 590.48.01. На момент начала экспериментов - примерно пару недель тому назад - версии vllm 0.16 ещё не было, но, как выяснилось в итоге, это не сильно повлияло на ситуацию. Основные костыли остались теми же самыми. Разве что кастомизация образа не для каждой модели нужна теперь. В целом там, понятное дело, никакого RocketScience нету (особенно после того, как почитаешь китайские форумы в поисках нюансов). Но если бы кто-то посидел заранее и собрал советы в одном месте - жизнь была бы немного проще )) поэтому делюсь. Итак, поехали.

https://habr.com/ru/articles/1006202/

#KimiK25 #DeepSeekv32 #GLM5 #Qwen35 #vllm #B200 #H200

Moonshot AI veröffentlicht Kimi K2.5 mit Agent Swarm Architektur. Das Open Source Modell (595 GB) koordiniert bis zu 100 Sub-Agenten parallel. Statt sequentieller Abarbeitung werden Teilaufgaben simultan gelöst. Das System unterstützt Vibe-Coding via visueller Analyse. Fokus liegt auf massiver Parallelisierung für Rechenzentren. #MoonshotAI #KimiK25 #AgentSwarm
https://www.all-ai.de/news/news26/agent-swarm-kimi2-5

Using well-chosen public sources, i let NotebookLM and Kimi K2.5 generate a presentation on a subject, amazing quality and very few errors. Good inspiration for the real presentation i will make myself. AI is not only creating "slop", we have to really start more research about how we incorporate AI in our work-processes in a responsible way, how we don't become "dumber", maintain skills, learning and human creativity 🤔
#AI #slop #learning #skills #creativity #NotebookLM #KimiK25

Using well-chosen public sources, i let NotebookLM and Kimi K2.5 generate a presentation on a subject, amazing quality and very few errors. Good inspiration for the real presentation i will make myself. AI is not only creating "slop", we have to really start more research about how we incorporate AI in our work-processes in a responsible way, how we don't become "dumber", maintain skills, learning and human creativity 🤔
#AI #slop #learning #skills #creativity #NotebookLM #KimiK25

Explore how the new Kimi K2.5 multimodal VLM leverages NVIDIA GPU‑accelerated endpoints, vLLM and the NeMo framework to deliver faster, open‑source vision‑language inference. The article walks through CUDA optimizations and practical deployment tips for AI researchers and developers. #KimiK25 #MultimodalVLM #NVIDIAGPU #NeMoFramework

🔗 https://aidailypost.com/news/build-kimi-k25-multimodal-vlm-nvidia-gpu-accelerated-endpoints

Explore how the new Kimi K2.5 multimodal VLM leverages NVIDIA GPU‑accelerated endpoints, vLLM and the NeMo framework to deliver faster, open‑source vision‑language inference. The article walks through CUDA optimizations and practical deployment tips for AI researchers and developers. #KimiK25 #MultimodalVLM #NVIDIAGPU #NeMoFramework

🔗 https://aidailypost.com/news/build-kimi-k25-multimodal-vlm-nvidia-gpu-accelerated-endpoints

Only just recently had some time to read the Kimi K2.5 technical report , an impressive open weights model (it's not fully opensource), the distance towards large paid proprietary foundation models is becoming very small.
https://github.com/MoonshotAI/Kimi-K2.5/blob/master/tech_report.pdf
It also feels that way in use:
https://www.kimi.com/chat
#AI #openweights #models #KimiK25 #China

Only just recently had some time to read the Kimi K2.5 technical report , an impressive open weights model (it's not fully opensource), the distance towards large paid proprietary foundation models is becoming very small.
https://github.com/MoonshotAI/Kimi-K2.5/blob/master/tech_report.pdf
It also feels that way in use:
https://www.kimi.com/chat
#AI #openweights #models #KimiK25 #China

Olen testannut uutta Kimi K2.5 -mallia, ja se tekee jotain, mihin Copilot, Gemini tai ChatGPT eivät tässä mittakaavassa pysty. Se ei vain kirjoita tekstiä, vaan luo valmiita tiedostoja, jotka ovat suoraan muokattavissa Microsoft Officella. Ja sen kyvyt ovat muutenkin aikamoisia, agenttipuolella erityisesti. Tsekkaa testini täältä 👉 https://youtu.be/TBttY7PHc-8

#Tekoäly #KimiK25

🎁 #KimiK25 Launch Top-Up Bonus: Get up to 50% in Vouchers! #MoonshotAI #AI #Promotion
💰 Tiered bonus on top-ups:
$100-$1,999 → 30%
$2,000-$9,999 → 40%
$10,000+ → 50% (max $10,000 voucher)
⏰ Until February 2, 2026 - voucher credited within 1 business day, valid for 30 days
🚀 Celebrating the launch of #KimiK25: Native multimodal agentic model with 1T parameters, 256K context & Agent Swarm with up to 100 parallel sub-agents

https://platform.moonshot.ai/docs/promotion#-kimi-k25-launch-top-up-bonus-get-up-to-50-in-vouchers

🎁 #KimiK25 Launch Top-Up Bonus: Get up to 50% in Vouchers! #MoonshotAI #AI #Promotion
💰 Tiered bonus on top-ups:
$100-$1,999 → 30%
$2,000-$9,999 → 40%
$10,000+ → 50% (max $10,000 voucher)
⏰ Until February 2, 2026 - voucher credited within 1 business day, valid for 30 days
🚀 Celebrating the launch of #KimiK25: Native multimodal agentic model with 1T parameters, 256K context & Agent Swarm with up to 100 parallel sub-agents

https://platform.moonshot.ai/docs/promotion#-kimi-k25-launch-top-up-bonus-get-up-to-50-in-vouchers

Moonshot just dropped a massive 595 GB Kimi K2.5 agent‑swarm model—one of the largest open‑source LLMs yet. It packs multimodal capabilities and aims to push AI scaling, while Reddit is calling for a leaner version. Curious how this could reshape inference and community‑driven AI? Dive in for the full breakdown. #MoonshotAI #KimiK25 #AgentSwarm #OpenSourceAI

🔗 https://aidailypost.com/news/moonshot-releases-595-gb-kimi-k25-agent-swarm-model-reddit-wants

Moonshot veröffentlicht Kimi K2.5 als Open-Source mit 1,04 Billionen Parametern. Das MoE-Modell ermöglicht Vibe Coding, also die Generierung von Code direkt aus Video-Inputs. Zusätzlich orchestriert es Agenten-Schwärme für komplexe Aufgaben. Die Gewichte liegen offen, was den Betrieb auf eigener Hardware erlaubt und technologisch zu US-Modellen aufschließt. #Moonshot #KimiK25 #OpenSource
https://www.all-ai.de/news/news26top/moonshot-kimi2-5

New open weights model Kimi K2.5
"self-directed agent swarm paradigm" ,
"For complex tasks, Kimi K2.5 can self-direct an agent swarm with up to 100 sub-agents, executing parallel workflows across up to 1,500 tool calls. Compared with a single-agent setup, this reduces execution time by up to 4.5x. The agent swarm is automatically created and orchestrated by Kimi K2.5 without any predefined subagents or workflow."
https://simonwillison.net/2026/Jan/27/kimi-k25/#atom-everything
#KimiK25 #openweights #AI

New open weights model Kimi K2.5
"self-directed agent swarm paradigm" ,
"For complex tasks, Kimi K2.5 can self-direct an agent swarm with up to 100 sub-agents, executing parallel workflows across up to 1,500 tool calls. Compared with a single-agent setup, this reduces execution time by up to 4.5x. The agent swarm is automatically created and orchestrated by Kimi K2.5 without any predefined subagents or workflow."
https://simonwillison.net/2026/Jan/27/kimi-k25/#atom-everything
#KimiK25 #openweights #AI

https://winbuzzer.com/2026/01/27/moonshot-ai-kimi-k2-5-open-source-model-xcxwbn/

Moonshot AI Unveils Kimi K2.5 Open-Source Model

#AI #OpenSourceAI #KimiK25 #MoonshotAI #AIAgents #AgenticAI #OpenSourceAI #AIModels #AICoding #ChinaAI

https://winbuzzer.com/2026/01/27/moonshot-ai-kimi-k2-5-open-source-model-xcxwbn/

Moonshot AI Unveils Kimi K2.5 Open-Source Model

#AI #OpenSourceAI #KimiK25 #MoonshotAI #AIAgents #AgenticAI #OpenSourceAI #AIModels #AICoding #ChinaAI