# gpt-oss Published articles for gpt-oss. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## MLPerf Inference v6.1: 5.7x Per-Accelerator Gains, a 512-GPU Run, and Vera Rubin's First Peer-Reviewed Numbers DevFeed: [MLPerf Inference v6.1: 5.7x Per-Accelerator Gains, a 512-GPU Run, and Vera Rubin's First Peer-Reviewed Numbers]() Original publisher: [Read original article]() Author: Harold Fritts Published: 2026-09-16T15:00:00Z Content type: news Language: en Sources: [StorageReview.com]() Topics: [Benchmark](), [Inference Performance](), [Retrieval Augmented Generation (RAG)](), [agentic-coding](), [GPU](), [Vera Rubin NVL72](), [Nvidia](), [Vera Rubin]() Tags: [agentic-coding](), [ai](), [benchmark](), [enterprise](), [gpt-oss](), [gpu](), [inference](), [llama-cpp](), [numbers](), [nvidia](), [qwen3](), [rag]() ### AI overview MLCommons published MLPerf Inference v6.1 with record participation, two new inference tests, and peer-reviewed results for several newly covered accelerators. The release reports a 5.7x improvement in the best per-accelerator DeepSeek-R1 server result compared with v5.1. ### Source excerpt MLCommons has published MLPerf Inference v6.1, and the round sets a participation record with 30 submitting organizations and 486 datacenter and edge results. Two new tests join the suite: an End-to-End RAG pipeline for the datacenter and an Edge Agentic Inference benchmark for single-user devices, and the results carry the first peer-reviewed numbers for NVIDIA's The post MLPerf Inference v6.1: 5.7x Per-Accelerator Gains, a 512-GPU Run, and Vera Rubin's First Peer-Reviewed Numbers appeared first on StorageReview.com. ## Quantization-Aware Healing: a compressed, 4-bit model that outperforms its full-precision original DevFeed: [Quantization-Aware Healing: a compressed, 4-bit model that outperforms its full-precision original]() Original publisher: [Read original article]() Author: Antonio Tiene; Iker García-Ferrero; Ali Hashemi; Bakbergen Ryskulov Published: 2026-08-25T11:39:24Z Content type: article Language: en Sources: [Hugging Face - Blog]() Topics: [AI Chat]() Tags: [architecture](), [benchmarks](), [blog](), [compression](), [fine-tuning](), [gpt](), [gpt-oss](), [hugging-face](), [llms](), [model](), [post-training](), [quantization](), [rlhf]() ### AI overview The article presents Quantization-Aware Healing (QAH), a method for recovering structurally compressed and 4-bit-quantized LLMs. It contrasts QAH with quantization-aware training and distillation, arguing that the latter can be limited when no independently trained full-precision version of the compressed architecture exists. ### Source excerpt A Blog post by Multiverse Computing on Hugging Face ## Jalapeño's first results show industry-leading speed and efficiency in AI inference DevFeed: [Jalapeño's first results show industry-leading speed and efficiency in AI inference]() Original publisher: [Read original article]() Published: 2026-08-25T07:00:00Z Content type: article Language: en Sources: [OpenAI News]() Topics: [Artificial Intelligence](), [Inference](), [Inference Performance](), [OpenAI](), [Hardware](), [Benchmark](), [gpt-oss](), [deepseek]() Tags: [ai](), [benchmark](), [deepseek](), [efficiency](), [engineering](), [gpt-oss](), [hardware](), [inference](), [latency](), [models](), [openai](), [performance](), [software](), [speed](), [systems]() ### AI overview OpenAI reports initial results for Jalapeño, its custom inference chip. The article says the chip delivers higher throughput, lower end-to-end latency, and greater AI work per watt across GPT-OSS 120B, DeepSeek R1, and Kimi K2.5 1T, based on tests using the InferenceX benchmark. ### Source excerpt Jalapeño is a custom inference chip from OpenAI that delivers faster, more power-efficient AI inference, with higher throughput and lower latency for modern models. ## Open-weight models are fast on Neon AI Gateway. Here's why DevFeed: [Open-weight models are fast on Neon AI Gateway. Here's why]() Original publisher: [Read original article]() Author: Carlota Soto Published: 2026-08-20T12:00:00Z Content type: article Language: en Sources: [Blog -- Neon Docs]() Topics: [databricks](), [Inference Performance](), [model-deployment](), [Low-Latency Inference](), [Caching](), [Mixture of Experts (MoE)](), [gpt-oss](), [Large Language Model]() Tags: [caching](), [databricks](), [gpt-oss](), [inference](), [mixture-of-experts-moe](), [optimization](), [performance](), [product](), [speed]() ### AI overview The article explains why open-weight models can deliver low latency on Neon AI Gateway when paired with Databricks Foundation Model APIs and an optimized serving stack. It highlights continuous batching, KV-cache paging, custom kernels, quantization, multi-GPU layouts, prompt caching, and Mixture-of-Experts architectures. ### Source excerpt Open-weight models are built for performance, but how they're served determines whether end users actually feel that speed. Neon AI Gateway calls Databricks Foundation Model APIs for Databricks-hosted open-weight models, a stack that has years of inference engineering behind it. ## How Reasoning Traces Work in Language Models DevFeed: [How Reasoning Traces Work in Language Models]() Original publisher: [Read original article]() Author: Armin Ronacher Published: 2026-08-19T00:00:00Z Content type: article Language: en Sources: [Armin Ronacher]() Topics: [Chain-of-thought](), [gpt-oss](), [Parser](), [API](), [Cache]() Tags: [ai](), [api](), [chain-of-thought](), [gpt-oss](), [reasoning](), [token]() ### AI overview The article explains reasoning traces as text emitted by a model into a scratchpad before its final answer. It discusses how GPT-OSS uses channel markers and a parser to route analysis into a separate stream, and argues that reasoning effort is shaped by system prompts and training rather than being solely a sampling-process property. ### Source excerpt A few weeks ago a paper was shared that showed how to extract reasoning traces from closed-weight models. Together with online discussions about tricking models into leaking them, it made me investigate it more out of curiosity. Twitter seems full of half-truths and confusion about how this works, so perhaps this helps some to understand what is happening. Hiding Traces Reasoning traces are usually hidden from us. We have lamented this, but mostly have to accept it. Open-weight models thankfully reveal them, and from their behavior you can see that their traces can be long and confusing. This is probably a good reason to separate them from what is normally shown to users. At minimum, UIs need to detect them. The industry has done a good job at making reasoning traces sound special and exotic, but they really are just text: the model is trained to emit its thinking into a scratchpad as part of its response, before its final answer. GPT-OSS's Harmony response format makes this easy to see: <|channel|>analysis<|message|> I need to work this out ... <|end|><|start|>assistant<|channel|>final<|message|> The answer is ... <|return|> The markers are special tokens, but the reasoning between them uses "the same text" as the final answer (just that GPT chain-of-thought text sounds really funny). When the model samples the analysis channel token, a parser routes the following text into a separate stream exposed through the Responses API. For closed models, presumably a simple model redacts and summarizes it. Reasoning Effort How much budget goes to reasoning? Earlier APIs exposed reasoning token budgets, making it seem like a property of the sampling process. In reality, reasoning effort is baked into the system prompt. GPT-OSS puts this into the system prompt: Reasoning: low That's it. Training produces the resulting behavior, such as emitting the token sequence that switches to the analysis channel. This also explains why changing the effort invalidates the KV cache. I think ## How Much Memory Does Your Agent Actually Need? DevFeed: [How Much Memory Does Your Agent Actually Need?]() Original publisher: [Read original article]() Author: Vatche Isahagian; Gaodan Fang; Jayaram Radhakrishnan; Punleuk Oum; Ashwath Vaithinathan Aravindan; Evelyn Duesterwald; G Thomas; Vinod Muthusamy; Merve Unuvar; Ayhan Sebin Published: 2026-08-18T18:09:38Z Content type: article Language: en Sources: [Hugging Face - Blog]() Topics: [Benchmark](), [Inference](), [long-context](), [gpt-oss](), [deepseek](), [Mixture of Experts (MoE)]() Tags: [agent](), [agentic](), [caching](), [deepseek](), [evaluation](), [gpt-oss](), [inference](), [memory](), [model](), [models](), [moe](), [performance](), [production](), [tokens]() ### AI overview ALTK-Evolve equips agents with agentic memory by distilling reusable guidelines from past trajectories and injecting them at inference time without weight updates or human annotation. An evaluation across eight models found that the effective amount of memory varies by model: stronger models benefited from full guideline sets, weaker models from compact cores plus task-specific retrieval, and saturated models showed no measurable improvement. For gpt-oss-120b, selective retrieval improved task completion by 16.1 percentage points with only a 5% token increase. ### Source excerpt Equipping an agent with agentic memory sounds simple: distill lessons from its past work, put them back in context, and more experience should mean better performance. It doesn't always work that way. When we scaled the evaluation to eight models -- from a 30B dense model to frontier proprietary systems -- one finding stood out: Agentic memory is not a feature you switch on. It's a dose you calibrate to the model. ## Making Knowledge Distillation Cheap Enough to Run at Scale DevFeed: [Making Knowledge Distillation Cheap Enough to Run at Scale]() Original publisher: [Read original article]() Author: Antonio Tiene; Iker García-Ferrero; Ali Hashemi; Bakbergen Ryskulov Published: 2026-08-10T10:05:36Z Content type: article Language: en Sources: [Hugging Face - Blog]() Topics: [LLM Techniques](), [Large Language Model](), [Post-training optimization](), [Model Routing](), [GPU](), [Machine learning](), [PyTorch](), [gpt-oss](), [Nvidia]() Tags: [blog](), [caching](), [compute](), [cost](), [gpt](), [gpt-oss](), [gpu](), [hugging-face](), [language-models](), [large-language-models](), [long-context](), [machine-learning](), [memory](), [performance](), [pytorch](), [qwen](), [research](), [training]() ### AI overview This article presents an efficient approach to knowledge distillation for large language models. It caches the teacher model's top-K logits offline and uses a fused, memory-efficient chunked KL-divergence loss, avoiding simultaneous teacher-student residency and full vocabulary-by-sequence-length matrices. The changes reduce VRAM use and training cost, enabling long-context distillation on a single GPU and making larger-scale experimentation more practical. ### Source excerpt A Blog post by Multiverse Computing on Hugging Face ## System Over Model, Tested: Reproducing Mythos's FreeBSD Find on Local Open-Weight Models DevFeed: [System Over Model, Tested: Reproducing Mythos's FreeBSD Find on Local Open-Weight Models]() Original publisher: [Read original article]() Author: clearbluejar Published: 2026-06-04T07:00:00Z Content type: article Language: en Sources: [clearbluejar]() Topics: [freebsd](), [gpt-oss](), [Artificial Intelligence](), [gemma4](), [bug](), [Large Language Model](), [Benchmark](), [Python]() Tags: [ai](), [benchmark](), [cve](), [false-positives](), [freebsd](), [gemma-4](), [gpt-oss](), [llm](), [llms](), [python](), [rce](), [vulnerability-research]() ### AI overview This article tests AISLE's vulnerability-finding pipeline on the local open-weight models gpt-oss-20b and gemma-4-31b-it using a FreeBSD RCE as the benchmark. The models initially appeared to miss the bug, but a rerun recovered it. The author reports that adding a reachability stage reduced false positives from 30 to 5 while retaining the CVE finding. ### Source excerpt Mythos found a 17-year-old FreeBSD RCE; AISLE reproduced it with gpt-5.4-nano via their nano-analyzer pipeline. I ran the pipeline on two local open-weight models, gpt-oss-20b and gemma-4-31b-it. The misses recovered on re-run. The real problem was the false-positive rate, and one extra system stage cut it from 30 to 5 with the CVE still standing. ## Sort providers by cost, latency, or throughput on AI Gateway DevFeed: [Sort providers by cost, latency, or throughput on AI Gateway]() Original publisher: [Read original article]() Author: Jerilyn Zheng Published: 2026-05-15T00:00:00Z Content type: release Language: en Sources: [Vercel News]() Topics: [Inference Performance](), [Routing (disambiguation)](), [Latency](), [gpt-oss](), [deepseek]() Tags: [ai](), [ai-gateway](), [deepseek](), [generation](), [gpt-oss](), [latency](), [model](), [models](), [routing](), [speed](), [tokens]() ### AI overview AI Gateway now supports sorting model providers by cost, time to first token, or tokens-per-second throughput. Ranking is calculated at request time, works with filtering and provider-order controls, and exposes routing decisions in response metadata. ### Source excerpt You can now sort the providers behind a model by cost, time to first token (TTFT), or throughput (TPS) in AI Gateway. The default provider order blends provider reliability, quality of model output, cost, and speed of response. You can now use sort for explicit control over ranking criteria. For models with many providers and noticeable cost or speed variation, you can use sort to optimize on your dimension of choice. Ranking is computed at request time, so newly added providers, price changes, and shifts in observed latency or throughput flow through automatically without any code changes. Set sort on providerOptions.gateway to one of the three values: Value Description Direction When to use 'cost' Sort by the provider's listed input price per million tokens Lowest price first High-volume, cost-sensitive work 'ttft' Sort by median time to first token, in ms Lowest latency first Latency-sensitive workloads where response speed matters 'tps' Sort by median tokens per second throughput Highest first Long-output generation where total response time matters most Basic usage Use sort to ensure optimizing for your metric of choice. In this example, AI Gateway has over five providers for GPT OSS 120B with different prices, so sorting by cost is a useful option for requests that want to route through the lowest price provider. Providers are tried in sort order. Fallback to the next provider only happens when the higher-ranked one is unavailable. Combine with other routing controls sort is compatible with other gateway routing options like Zero Data Retention (ZDR). The example below uses deepseek/deepseek-v4-pro for an interactive request where latency and data retention matter: AI Gateway filters to only providers for Deepseek V4 Pro that have zero data retention, and then sorts the remaining providers by time to first token (TTFT). sort also composes with order: providers listed in order are promoted to the front, and the remaining providers follow the requested sort crit ## Helping developers build safer AI experiences for teens DevFeed: [Helping developers build safer AI experiences for teens]() Original publisher: [Read original article]() Published: 2026-03-24T11:00:00Z Content type: article Language: en Sources: [OpenAI News]() Topics: [Artificial Intelligence](), [gpt-oss](), [AI Strategy](), [OpenAI](), [releases](), [model-deployment]() Tags: [ai](), [building](), [developers](), [ecosystem](), [gpt-oss](), [model](), [models](), [open](), [openai](), [releases](), [safety](), [support](), [tools](), [work]() ### AI overview OpenAI is releasing prompt-based teen safety policies for developers using the open-weight gpt-oss-safeguard model. The policies help translate teen-specific safety requirements into operational classifiers and support safer, more consistent deployment of AI systems. ### Source excerpt OpenAI releases prompt-based teen safety policies for developers using gpt-oss-safeguard, helping moderate age-specific risks in AI systems. ## IBM and UC Berkeley Diagnose Why Enterprise Agents Fail Using IT-Bench and MAST DevFeed: [IBM and UC Berkeley Diagnose Why Enterprise Agents Fail Using IT-Bench and MAST]() Original publisher: [Read original article]() Author: Ayhan Sebin; Rohan Arora; Saurabh Jha Published: 2026-02-18T16:15:45Z Content type: article Language: en Sources: [Hugging Face - Blog]() Topics: [Benchmark](), [ibm](), [Large Language Model](), [site-reliability-engineering](), [incident](), [Kubernetes](), [finops](), [Security]() Tags: [agentic](), [agents](), [analysis](), [automation](), [benchmarks](), [enterprise](), [evaluation](), [gpt-oss](), [hallucinations](), [ibm](), [incident](), [kubernetes](), [llm](), [logs](), [loops](), [metrics](), [performance](), [research](), [systems](), [tool]() ### AI overview IBM Research and UC Berkeley analyze why agentic LLM systems fail in enterprise IT automation using ITBench traces and the MAST failure taxonomy. Their analysis of 310 SRE traces compares Gemini-3-Flash, Kimi-K2, and GPT-OSS-120B, identifying verification errors, cascading failures, premature termination, looping, and hallucinations as major reliability problems. ### Source excerpt IBM Research and UC Berkeley collaborated to study how agentic LLM systems break in real-world IT automation, for tasks involving incident triage, logs/metrics queries, and Kubernetes actions in long-horizon tool loops. Benchmarks typically reduce performance to a single number, telling you whether an agent failed but never why. To solve this black-box problem, we applied MAST (Multi-Agent System Failure Taxonomy), an emerging practice for diagnosing agentic reliability ). ## Unlocking Agentic RL Training for GPT-OSS: A Practical Retrospective DevFeed: [Unlocking Agentic RL Training for GPT-OSS: A Practical Retrospective]() Original publisher: [Read original article]() Author: Jason Zhu; Hejian Sang; Arup De; Rohit Jain; Yanning Chen Published: 2026-01-27T01:53:15Z Content type: article Language: en Sources: [Hugging Face - Blog]() Topics: [Artificial Intelligence](), [gpt-oss](), [Reinforcement learning](), [agentic-coding](), [Retool](), [Frameworks](), [qwen](), [coding](), [hugging face]() Tags: [agentic](), [agentic-coding](), [agents](), [ai](), [applications](), [blog](), [code](), [coding](), [company](), [fine-tuning](), [gpt-oss](), [hugging-face](), [linkedin](), [models](), [open-source](), [qwen](), [reasoning](), [reinforcement-learning](), [rl](), [training]() ### AI overview LinkedIn's retrospective describes experiments to train GPT-OSS models for agentic reinforcement learning. It covers tool interaction, Harmony chat-template support, rollout and tool parsing, ReTool coding tasks, attention-sink fixes, and benchmark results using GPT-OSS-20B, GPT-OSS-120B, and Qwen-2.5-32B. ### Source excerpt LinkedIn is an AI-first company that's built agents to help professionals be more successful. In this setting, models must reason over incomplete information, interact with structured services, and adapt to evolving user intent across multiple steps rather than produce a single static response. ## gpt-oss-safeguard technical report DevFeed: [gpt-oss-safeguard technical report]() Original publisher: [Read original article]() Published: 2025-10-29T00:00:00Z Content type: article Language: en Sources: [OpenAI News]() Topics: [gpt-oss](), [Chain-of-thought](), [Open Source](), [API]() Tags: [api](), [chain-of-thought](), [gpt-oss](), [open-source](), [oss](), [reasoning](), [report](), [safety](), [technical]() ### AI overview This technical report introduces gpt-oss-safeguard-120b and gpt-oss-safeguard-20b, open-weight reasoning models fine-tuned from gpt-oss to classify content according to a supplied policy. It describes their capabilities, customization, chain-of-thought support, Responses API compatibility, and baseline safety evaluations. ### Source excerpt gpt-oss-safeguard-120b and gpt-oss-safeguard-20b are two open-weight reasoning models post-trained from the gpt-oss models and trained to reason from a provided policy in order to label content under that policy. In this report, we describe gpt-oss-safeguard's capabilities and provide our baseline safety evaluations on the gpt-oss-safeguard models, using the underlying gpt-oss models as a baseline. For more information about the development and architecture of the underlying gpt-oss models, see the original gpt-oss model model card⁠. ## Introducing gpt-oss-safeguard DevFeed: [Introducing gpt-oss-safeguard]() Original publisher: [Read original article]() Published: 2025-10-29T00:00:00Z Content type: release Language: en Sources: [OpenAI News]() Topics: [gpt-oss](), [OpenAI](), [Chain-of-thought](), [Inference](), [hugging face](), [model-deployment]() Tags: [chain-of-thought](), [developers](), [gpt-oss](), [hugging-face](), [inference](), [models](), [open](), [openai](), [oss](), [product](), [reasoning](), [research](), [safety]() ### AI overview OpenAI introduces gpt-oss-safeguard, a research preview of open-weight reasoning models for safety classification. The models let developers provide and revise their own policies at inference time, classify messages and conversations, and review the model's reasoning. They are available in 120b and 20b sizes under the Apache 2.0 license and can be downloaded from Hugging Face. ### Source excerpt OpenAI introduces gpt-oss-safeguard--open-weight reasoning models for safety classification that let developers apply and iterate on custom policies. ## Google Cloud C4 Brings a 70% TCO improvement on GPT OSS with Intel and Hugging Face DevFeed: [Google Cloud C4 Brings a 70% TCO improvement on GPT OSS with Intel and Hugging Face]() Original publisher: [Read original article]() Author: Jiqing.Feng; Matrix Yao; Ke Ding; Ilyas Moutawwakil Published: 2025-10-16T00:00:00Z Content type: article Language: en Sources: [Hugging Face - Blog]() Topics: [moe](), [cpu](), [AI, ML & Data Engineering]() Tags: [cpu](), [google-cloud](), [gpt-oss](), [hugging-face](), [inference](), [intel](), [llm](), [mixture-of-experts-moe](), [open-source](), [performance]() ### AI overview The article benchmarks GPT OSS mixture-of-experts text generation on Google Cloud Intel Xeon virtual machines. It describes an expert-execution optimization and reports throughput, latency, and total-cost-of-ownership comparisons between Xeon generations. ### Source excerpt We're on a journey to advance and democratize artificial intelligence through open source and open science. ## Nemotron-Personas-Japan: ソブリン AI のための合成データセット DevFeed: [Nemotron-Personas-Japan: ソブリン AI のための合成データセット]() Original publisher: [Read original article]() Author: Atsunori Fujita; Masaya Ogushi; Vincent Gong; Kotaro Yamamoto; Yoshi Suhara; Dane Corneil; Yev Meyer Published: 2025-09-26T06:25:50Z Content type: article Language: ja Sources: [Hugging Face - Blog]() Topics: [datasets](), [Nemotron](), [Artificial Intelligence](), [data](), [Large Language Model](), [NeMo](), [Nvidia](), [gpt-oss](), [pii](), [Pydantic]() Tags: [ai](), [apache](), [data](), [gpt-oss](), [japan](), [llm](), [nemo](), [nemotron](), [nvidia](), [oss](), [pii]() ### AI overview NVIDIA has released Nemotron-Personas-Japan, an open synthetic dataset of Japanese personas designed to support culturally grounded and privacy-preserving AI development. Built with NeMo Data Designer, it contains six million Japanese-language personas aligned with Japanese demographic, geographic, cultural, and labor statistics, without including personally identifiable information. ### Source excerpt 高品質で多様なトレーニングデータなしに、日本文化を真に理解するAIを構築することはこれまでほぼ不可能でした。これを変えるため、NVIDIAは、日本の人口統計、地理的分布、文化的特性に沿ったペルソナを含む初のオープン合成データセット、Nemotron-Personas-Japan を公開しました。CC BY 4.0 ライセンスのもと提供される本データセットは、機微な個人データに依存することなく日本社会を反映した AI システム構築のための、プライバシー保護と規制対応を両立した基盤を提供します。 NVIDIA のエンタープライズ向け合成データ生成システム、NeMo Data Designer を用いて作成されたNemotron-Personas-Japan は、すでに広く利用されている US Personas データセットの成功を機に日本版として開発されました。本リリースは、各国・地域におけるソブリン AI 開発を支援する合成ペルソナデータセットとデータ構築方法のグローバルコレクションの第一弾です。 ## Scaleway on Hugging Face Inference Providers 🔥 DevFeed: [Scaleway on Hugging Face Inference Providers 🔥]() Original publisher: [Read original article]() Author: Guillaume Noale; Franck Pagny; Fred Bardolle; Guillaume Calmettes; Constance Morales; Célina Hanouti; Julien Chaumond; Simon Brandeis; Lucain Pouget Published: 2025-09-19T00:00:00Z Content type: article Language: en Sources: [Hugging Face - Blog]() Topics: [hugging face](), [scaleway](), [Inference](), [inference-providers](), [Serverless](), [gpt-oss](), [SDKs](), [API keys](), [Latency](), [Sovereign AI]() Tags: [api-keys](), [enterprise](), [function-calling](), [gemma](), [gpt-oss](), [hub](), [hugging-face](), [inference](), [inference-providers](), [llms](), [low-latency](), [multimodal](), [open](), [open-source](), [partnerships](), [providers](), [qwen3](), [scaleway](), [sdks](), [serverless](), [text-generation]() ### AI overview Hugging Face announces Scaleway as a supported Inference Provider on the Hugging Face Hub. The integration provides serverless access to open-weight and frontier AI models through model pages, client SDKs, and APIs, with European data centers, pay-per-token pricing, low latency, and production-oriented features. ### Source excerpt We're on a journey to advance and democratize artificial intelligence through open source and open science. ## Tricks from OpenAI gpt-oss YOU 🫵 can use with transformers DevFeed: [Tricks from OpenAI gpt-oss YOU 🫵 can use with transformers]() Original publisher: [Read original article]() Author: Aritra Roy Gosthipaty; Sergio Paniego; Vaibhav Srivastav; Pedro Cuenca; Arthur Zucker; Nathan Habib; Cyril Vallez Published: 2025-09-11T00:00:00Z Content type: article Language: en Sources: [Hugging Face - Blog]() Topics: [Transformers](), [gpt-oss](), [Mixture of Experts (MoE)](), [CUDA](), [Large Language Model](), [OpenAI](), [PyTorch](), [Open Source](), [C++](), [Python]() Tags: [c-plus-plus](), [cuda](), [gpt-oss](), [guide](), [llms](), [mixture-of-experts-moe](), [open-source](), [openai](), [optimization](), [python](), [pytorch](), [transformers]() ### AI overview This article explains performance improvements in the Transformers toolkit, including specialized kernels, kernel fusion, Flash Attention, and pre-built kernel binaries. It highlights how GPT-OSS, a Mixture of Experts model, uses custom kernels distributed through the Hub. ### Source excerpt We're on a journey to advance and democratize artificial intelligence through open source and open science. ## Heroku AI Expands Model Offering with OpenAI's gpt-oss-120b DevFeed: [Heroku AI Expands Model Offering with OpenAI's gpt-oss-120b]() Original publisher: [Read original article]() Author: Anush DSouza Published: 2025-08-20T15:00:14Z Content type: release Language: en Sources: [Heroku]() Topics: [Heroku AI](), [gpt-oss](), [OpenAI](), [Heroku](), [API](), [Mixture of Experts (MoE)](), [SDKs](), [Open Source Models & Datasets]() Tags: [agents](), [ai](), [ai-infrastructure](), [ai-machine-learning](), [api](), [deploy](), [engineering](), [function-calling](), [gpt-oss](), [heroku](), [heroku-ai](), [managed-inference-and-agents](), [mixture-of-experts-moe](), [open-source](), [openai](), [python](), [sdk]() ### AI overview Heroku announces that OpenAI's open-weight gpt-oss-120b model is available through Heroku Managed Inference and Agents. Developers can access it through an OpenAI-compatible chat completions API and use it to build and deploy AI applications and agentic workflows. ### Source excerpt Start building with OpenAI's new open-weight model, gpt-oss-120b, now available on Heroku Managed Inference and Agents. This gives developers a powerful, transparent, and flexible way to build and deploy AI applications on the platform they already trust. Access gpt-oss-120b with our OpenAI-compatible chat completions API, which you can drop into any OpenAI-compatible SDK or framework. [...] The post Heroku AI Expands Model Offering with OpenAI's gpt-oss-120b appeared first on Heroku. ## GPT-5 and the shift toward the agentic web DevFeed: [GPT-5 and the shift toward the agentic web]() Original publisher: [Read original article]() Published: 2025-08-08T00:00:00Z Content type: opinion Language: en Sources: [Tim Kellogg]() Topics: [Large Language Model](), [Artificial Intelligence](), [Claude Code](), [agent observability]() Tags: [ai](), [ai-agents](), [claude-code](), [gpt](), [gpt-oss](), [llm]() ### AI overview This commentary argues that GPT-5 should be evaluated less by whether it feels dramatically better as a language model and more by how effectively it supports an agentic web. It highlights software harnesses, tool use, reinforcement learning, and features such as Claude Code's subagents as important parts of that shift. ### Source excerpt the best way to emphasize the importance of this week's developments is to go all the way back to January and see how we got here. ## Welcome GPT OSS, the new open-source model family from OpenAI! DevFeed: [Welcome GPT OSS, the new open-source model family from OpenAI!]() Original publisher: [Read original article]() Author: Vaibhav Srivastav; Pedro Cuenca; Lewis Tunstall; Clem 🤗; Matthew Carrigan; Clémentine Fourrier; Célina Hanouti; Lucain Pouget; Marc Sun; Simon Pagezy Published: 2025-08-05T00:00:00Z Content type: release Language: en Sources: [Hugging Face - Blog]() Topics: [gpt-oss](), [Open Source](), [OpenAI](), [Artificial Intelligence](), [inference-providers](), [llama.cpp](), [moe](), [Ollama](), [Python](), [quantization](), [Transformers](), [vllm]() Tags: [api](), [artificial-intelligence](), [community](), [gpt](), [gpt-oss](), [inference-providers](), [javascript](), [llama-cpp](), [llm](), [moe](), [ollama](), [open-source](), [openai](), [python](), [quantization](), [transformers](), [vllm]() ### AI overview Hugging Face welcomes OpenAI's gpt-oss open-source model family. The article describes the models' Apache 2.0 licensing, local deployment options, reasoning and tool-use capabilities, MoE architecture, quantization, supported inference implementations, and access through Inference Providers and the Responses API. ### Source excerpt We're on a journey to advance and democratize artificial intelligence through open source and open science.