RT @PawelHuryn: Beats Sonnet 4.6 on graduate-level reasoning. 4B active parameters. Runs on a 24GB Mac Mini. Gemma 4's 26B model scores 82.3% on GPQA Diamond — vs Sonnet 4.6's 74%. It's a mixture-of-experts that activates only 4B parameters per inference. Apache 2.0. The 31B variant goes further: 84.3% on the same benchmark. An open source model outperforming the current frontier on graduate-level reasoning. Sonnet 4.6 still wins on agentic coding (SWE-bench 79.6%). But frontier-level reasoning now runs locally, on your hardware, for free. Google AI (@GoogleAI) Today, we’re launching Gemma 4, our most intelligent open models to date. Built with the same breakthrough technology as Gemini 3, Gemma 4 brings advanced reasoning to your personal hardware and devices. Here’s what Gemma 4 unlocks for developers: — Intelligence-per-parameter: Our 31B (Dense) and 26B (MoE) models deliver state-of-the-art performance for their size, outcompeting models 20x their size on @arena — Commercial flexibility: Released under a permissive Apache 2.0 license for complete developer flexibility and digital sovereignty — Agentic workflows: Native support for function-calling and structured JSON output allows you to build reliable, autonomous agents — Multimodal edge AI: The E2B and E4B models bring native vision, audio, and low latency to mobile and IoT devices — Long-context reasoning: Up to 256K context windows allow you to process entire repositories or large documents in a single prompt Whether you're building global applications in 140+ languages or local-first AI code assistants, Gemma 4 is built to be your foundation. Explore in @GoogleAIStudio or download the weights on @HuggingFace, @Kaggle, and @Ollama. Video — https://nitter.net/GoogleAI/status/2039735543068504476#m

Mehr auf Arint.info

#Apache #Gemini #global #Google #HuggingFace #nitter #Ollama #opensource #SWE #arint_info

https://x.com/PawelHuryn/status/2039781705884590326#m

Greetings in ǃGãǃne!

https://ai.forfun.su/2026/04/04/greetings-in-%c7%83ga%c7%83ne/

This is hilarious. There is a site that does the whole exposé on how #ClaudeCode works.

https://ccunpacked.dev/

They should have called it CUCK: Claude Unpacked Code Knowledge.

Because that's is what Anthropic is going to feel the next coming weeks.

#Programming #Programmers #Coding #Code #SoftwareDevelopment #WebDevelopment #WebDev #AppDevelopment #CLI #Linux #FOSS #OSS #OpenClaw #Claude #Codex #Llama #Ollama #LlamaCCP #LLM #LargeLanguageModel #AI #LMStudio

Tested Cogito V1 8B on my Linux server. 83 t/s, 5.4GB VRAM, 131k context. The real story is where it deliberately wrote worse code because it decided a beginner needed simplicity over efficiency -- and admitted it! That's IDA self-reflection making a live call.
I guess a 5GB model with a conscience is worth more than a 70B model with none?

Read the full breakdown below.

#LocalAI #Ollama #HomeLabAI #LLM #AIBenchmark

https://goarcherdynamics.com/2026/04/03/aihome-cogito-v1-8b-review/?utm_source=mastodon&utm_medium=jetpack_social

Обзор расширения Code Assistant от Яндекс для IDE VSCode

Столкнулся с проблемой оплаты зарубежного агента LLM и решил сделать задачу отечественным инструментом, в нем много изменилось с выхода в 2024 году. Вам решать юзать или нет, пишу дабы сократить время на настройку кто кинет взор как и я. С установкой нового расширения с адреса https://sourcecraft.dev/me/codeassistant/settings в нем визуально поменялось все кардинально: появился чат, diff, настройки правил и скилов но самое интересно, что не дало мне удалить его сразу это индексация проекта в эмбеддингах, конечно не cursor в котором по слухам проект индексируется в виде графов и храниться в оперативном облаке, но уже что то.

https://habr.com/ru/articles/1018644/

#vscode #ai_agent #ollama #code_assistant #sourcecraft

https://note.com/kazu_t/n/n8ab40b7435ff#108d9ff0-ac20-496e-a9a2-1612d2338e97

#ai
#ollama

Ich bin gerade dabei auf #Linux umzusteigen, und habe mir meinen eigenen Home Server eingerichtet. Jetzt kann ich endlich meine Fotos in der privaten Cloud mit #immich katalogisieren. Aber eine Sache, die mir richtig gute gefällt und die ich nicht auf dem Schirm hatte, ist die Möglichkeit ein lokales #AI Modell für Suche nach Bildern zu nutzen. Es macht echt Spaß mit Texteingabe Bilder zu finden, die ich vergessen (oder verdrängt) hatte.

Und jetzt frage ich mich, ginge sowas nicht auch mit #Email? Ich habe mal nach #Thunderbird plugins gesucht, und da gibt es welche mit #ollama Support - aber nur für Schreiben, Übersetzen und Zusammenfassen einzelner Emails. Da kann ich sie auch gleich selber alle durchlesen... Ich will "Sitzung bei XY mit Thema ABC" eingeben können und Treffer erhalten, die diese bestimmten Wörter nicht direkt benutzen - und das alles mit voller Kontrolle über die digitale #Privatsphäre. Das wäre mal wirklich hilfreich

I've been playing with #Ollama recently-- now that they have new Qwen MLX models my base level Mac Studio can easily run, I'm more excited about AI than I've been recently.

And all of a sudden, I realized why: the future that I feel like has been presented to us for a while is “Corporate overlord shitheads will own the means of computation and you will pay them whatever they want to participate in society”

But now it's like, “oh, nope, sorry, _I_ am going to own the means of computation”