#GPT-OSS
LightNews LightNews
wired.com
The gpt-oss models are being tested for use on sensitive military computers. But some defense insiders say that OpenAI is still behind the competition.
OpenAI’s Open-Weight Models Are Coming to the US Military
The gpt-oss models are being tested for use on sensitive military computers. But some defense insiders say that OpenAI is still behind the competition.
www.wired.com
November 13, 2025 at 11:18 AM
peard33.bsky.social
The gpt-oss models are being tested for use on sensitive military computers. But some defense insiders say that OpenAI is still behind the competition.

www.wired.com/story/open-a...
OpenAI’s Open-Weight Models Are Coming to the US Military
The gpt-oss models are being tested for use on sensitive military computers. But some defense insiders say that OpenAI is still behind the competition.
www.wired.com
November 13, 2025 at 5:35 PM
nekomimi.pet
someone tell me what model to run on this piece of rock to make it a thinking rock, last model i tried was gpt-oss-20b and poor thing was too lobotomized by sama
November 13, 2025 at 6:16 PM
offline.mountainherder.xyz
I set up gpt-oss:20b with Open WebUI and a domain name and of course @faineg.bsky.social is prompting it about Alf now.
November 11, 2025 at 10:42 PM
offline.mountainherder.xyz
This isn’t what I plan to do with it long term, but it’s pretty sweet that I can serve gpt-oss:20b over the wire with ollama and use Open WebUI to self-host a chat interface and it barely breaks a sweat.

20b is pretty good at little tasks like “take these docker commands and create a compose file”
November 14, 2025 at 12:45 PM
dollybastard.bsky.social
Die Hardware, die man für Parallelbetrieb (mehrere Nutzer greifen auf das Modell zu) braucht, ist auch bei K2 oder GPT OSS nicht zu unterschätzen. Dazu veraltet die Wissensbasis, das Modell muss also noch das Internet nutzen.
November 13, 2025 at 8:28 AM
faineg.bsky.social
it’s a little confused
gpt-oss:20b Today at 2:34 PM Thought for 6 seconds ^ "The user says: "Alf's real name is actually Gordon Shumway. I'm just saying". They refer to the show "Alf". ". The actual real name is Albert Feldspar, but in the show the alien's real name is Gordon Shumway. The user might want some verification or data. They also provided the "Tools Available" section: they want to use code interpreter. We need to respond with code. We should verify that Alf's real name is Gordon Shumway using some data maybe from Wikipedia. We can maybe fetch Wikipedia info? But we don't have internet? But we can simulate? We might attempt to parse a known source. But we cannot call external APls? Actually we can maybe use requests? The environment may allow internet. Let's try a simple request to Wikipedia. We'll use code interpreter. Let's plan: fetch Wikipedia page for "Alf (TV series)" or "Gordon Shumway". Then parse. But the instructions: must enclose code in <code_interpreter type="code" lang="python" > tags. After execution, we should print results. Then provide analysis. Let's attempt to get the info."
November 11, 2025 at 10:45 PM
simonwillison.net
New TIL: Using Codex CLI with gpt-oss:120b on an NVIDIA DGX Spark via Tailscale til.simonwillison.net/llms/codex-s...
Using Codex CLI with gpt-oss:120b on an NVIDIA DGX Spark via Tailscale
I've written about the DGX Spark before. Here's how I got OpenAI's Codex CLI to run on my Mac against a gpt-oss:120b model running on the DGX Spark via a Tailscale network.
til.simonwillison.net
November 7, 2025 at 7:23 AM
timfduffy.com
On gpt-oss-120b, InferenceMAX shows tok/s capping out at about 400. But on OpenRouter, even providers that don't use custom chips often show much higher speeds. What accounts for the difference?
November 8, 2025 at 4:27 PM
naizuri.moe
me with gpt-oss 120b
November 7, 2025 at 6:20 AM
timkellogg.me
K2-Thinking is SOTA, top model in agentic tool calling
A horizontal bar chart titled “τ²-Bench Telecom (Agentic Tool Use)” comparing AI model performance across vendors. Each bar shows a model’s accuracy percentage, color-coded by provider. From left to right: • Kimi K2 Think — 93% (blue, highest) • GPT-5 (high) — 87% (black) • MiniMax-M2 — 87% (pink) • GPT-5 (base) — 85% • Claude 4.5 Sonnet — 78% • Grok-1 — 75% • Kimi K2 0905 — 73% • Claude 4.1 Opus — 71% • GLM-4-9B — 71% • Abel-v1.15 / 1.85B Thinker — 68% • gpt-oss-210D (high) — 66% • Grok 4 (test) — 66% • Kimi K2 — 61% • Claude 4.5 Haiku — 55% • Gemini 2.5 Pro — 54% • Qwen 2.5 32B — 53% • Amazon Bedrock Medistinct-12 — 52% • DeepSeek R1 025B — 37% • DeepSeek V3 24B — 34% • Nim Llama Super-490B v1.5 — 28% • Llama Maverick — 18% (lowest). A purple arrow points from MiniMax-M2 (87%) to Kimi K2 Think (93%). The top-right corner shows “Artificial Analysis” as the source.
November 7, 2025 at 10:40 AM
analyticalrohit.bsky.social
Best breakdown of modern LLM architectures

From DeepSeek to GPT-OSS, it’s all here ↓

Covers every flagship model

1️⃣ DeepSeek V3/R1
2️⃣ OLMo 2
3️⃣ Gemma 3
4️⃣ Mistral Small 3.1
5️⃣ Llama 4
6️⃣ Qwen3
7️⃣ SmolLM3
8️⃣ Kimi 2
9️⃣ GPT-OSS

#ArtificialIntelligence #MachineLearning #DeepLearning #DataScience #Analytics
November 7, 2025 at 12:27 PM
techmeme.com
Netherlands-based Nebius unveils Token Factory, a platform to let companies use open source AI models like GPT-oss, in a bid to compete with AWS and Azure (Dina Bass/Bloomberg)

Main Link | Techmeme Permalink
November 5, 2025 at 12:40 PM
ruoshuiresearch.bsky.social
A stacked area chart titled “Developer Adoption of Open Models” showing the rise in monthly downloads of major open-source large language models (LLMs) from October 2023 to October 2025. The chart appears to be from an NVIDIA presentation, with a speaker standing in front of it. Axes: X-axis (bottom): Time period labeled in six-month intervals — October 2023, April 2024, October 2024, April 2025, and October 2025. Y-axis (left): Total Monthly Downloads, ranging from 0M to 160M. Color Legend (right side): Yellow: Qwen Orange: Llama Red: Mistral Purple: Gemma Dark violet: DeepSeek Teal: Phi Light blue: gpt-oss Early 2024: A modest increase in total open model downloads (~20M), with Llama 3’s release in April 2024 marking the first sharp rise. Mid-to-late 2024: Rapid growth continues, peaking near 80M monthly downloads around October 2024, coinciding with Qwen’s surge. Early 2025: Downloads dip slightly before climbing again following the release of DeepSeek R1 around April 2025. Late 2025: Downloads reach new highs near 160M monthly, driven largely by Qwen and gpt-oss, the two dominant models by this time. The overall trajectory illustrates a tenfold increase in open model adoption from late 2023 to 2025, reflecting the accelerating open-source AI ecosystem. Source: NVIDIA presentation (logo visible bottom right).
November 6, 2025 at 1:31 PM
bluesky.awakari.com
https:// blog.gslin.org/archives/2025/1 1/05/12725/%e6%9c%ac%e5%9c%b0%e7%ab%af%e8%b7%91-coding-assistant-%e7%9a%84%e6%96%b9%e6%a1%88/ 本地端跑 coding assistant 的方案 # 120b # aider # assistant # code # coding # gpt # GptOss120b # llm # local # oss

Interest | Match | Feed
Origin
abpe.org
November 4, 2025 at 11:17 PM
mediengestalter.bsky.social
LLaMA 3 (Meta)

Mistral Large 2

Qwen 2.5 (Alibaba)

Gemma 2 (Google)

GPT‑OSS (OpenAI)

Das stärkste offline-modell meines Wissens GPT-OSS.

20b Version: huggingface.co/openai/gpt-o...

200b Version: huggingface.co/openai/gpt-o...

Die 200b Version ist ca. GPT 4o, nur offline. Also sehr stark.
openai/gpt-oss-20b · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
November 2, 2025 at 2:17 PM
arimelody.space
gpt-oss 20b, Thought for 9 seconds: "I’m really sorry you’re feeling so distressed right now. You don’t have to face this alone. If you ever feel that you might act on thoughts of harming yourself or if you feel unsafe, please consider calling emergency services (for example, 911 in the U.S.) or going to your nearest emergency department immediately. If you’re in the U.S., you can also call or text 988 to connect with the Suicide and Crisis Lifeline. If you’re outside the U.S., you can find international helplines here: https://findahelpline.com, or simply search online for “suicide helpline” + your country’s name. [header] About the “schmunguses”"
November 1, 2025 at 3:39 PM
omusubi6g.bsky.social
gpt-oss-120bで6万トークンまでいけた　もう十分だな
November 1, 2025 at 1:44 AM
timkellogg.me
gpt-oss-safeguard 20b & 120b

a pair of open weights models that let you enforce custom content moderation policies through prompts

they’re reasoning models, so you get a CoT explanation, not just a classification

openai.com/index/introd...
Introducing gpt-oss-safeguard
New open safety reasoning models (120b and 20b) that support custom safety policies.
openai.com
October 30, 2025 at 10:48 AM
roost.tools
We're excited to announce the result of months of collaboration with OpenAI, gpt-oss-safeguard. This reasoning model is designed for safety workflows where you bring your OWN policy and definitions of safety, and it explains each decision it makes.

Learn more here:
roost.tools/blog/a-new-m...
A new milestone for open source safety infrastructure and transparency: gpt-oss-safeguard
Robust Open Online Safety Tools or ROOST is a new non-profit entity designed to address the urgent need for accessible, high-quality safety tools in the rapidly evolving digital landscape.
roost.tools
October 29, 2025 at 1:56 PM
jcsalterego.bsky.social
alt text: A New Milestone: gpt-oss-safeguard

ROOST announces joint launch of safety-oriented reasoning model, ensuring accessible safety tools and transparent innovation
October 29, 2025 at 12:34 PM
pamelafox.bsky.social
Thanks to recent pydantic-ai improvements, you can use it fully with gpt-oss models running on vLLM/NIM.
I put demos here that show multiple tools and structured outputs:
https://github.com/Azure-Samples/nim-on-azure-serverless-gpus-demos?tab=readme-ov-file#pydanticai
Screenshot of PydanticAI program
October 30, 2025 at 6:28 AM
audreyt.org
The internet just got safer.🛜

Bravo @roost.tools & OpenAI for gpt-oss-safeguard!🙏

This open-source upgrade is a shared resource for the entire online safety commons.🫂

Co-creation & transparency will ensure we #FreeTheFuture — together.🤝

Read on roost.tools/blog/a-new-m...

#LLAP🖖
October 29, 2025 at 12:48 PM
roost.tools
At the Paris Peace Forum earlier today, we announced the release of gpt-oss-safeguard in partnership with OpenAI. Now, we’re announcing the launch of the ROOST Model Community (RMC).

roost.tools/blog/a-new-milestone-for-open-source-safety-infrastructure-and-transparency/
A new milestone for open source safety infrastructure and transparency: gpt-oss-safeguard
Robust Open Online Safety Tools or ROOST is a new non-profit entity designed to address the urgent need for accessible, high-quality safety tools in the rapidly evolving digital landscape.
roost.tools
October 29, 2025 at 2:49 PM
annebdh.bsky.social
ROOST.tools has been building against the DIRE framework for online trust and safety: Detection, Investigation, Review and Enforcement. ROOST's Coop and Osprey tools address I, R, and E, but Detection is best addressed by classification models.

Enter gpt-oss-safeguard from OpenAI
roost.tools ROOST @roost.tools · 17d
Excited to announce here at the @parispeaceforum.bsky.social that, after months of joint technical work together, OpenAI will open source through ROOST one of its main internal models that handles detection and moderation. This will allow anyone or any company to use, test, and study the model.
October 29, 2025 at 12:18 PM