Skip to main content
Back

OpenAI Finally Releases Open Models Since GPT-2

New open-weight models, GitHub launches free inference API, Anthropic upgrades Claude Code, and more..

August 5, 2025 4 min read

Hey,

Welcome to AlphaSignal, the most read source of news by AI engineers and researchers. Every day, we identity and summarize the top 1% of news, papers, models, and repos, so you’re always up to date.


Here’s today’s roundup:

./SUMMARY

Read time: 4 min 19 sec

Top News

▪ OpenAI Releases First Open-Weight Models Since GPT-2

Lambda

▪ Skip manual setup, launch NVIDIA GPU clusters fast via Lambda

Trending Signals

▪ OpenAI improves ChatGPT to support real-world goals, not screen time

▪ Hugging Face shares a guide on LLM scaling techniques

▪ GitHub makes its inference API free for all users

▪ Google expands LLM evaluation with Kaggle-based open strategy game competitions

▪ Anthropic upgrades Claude Code with improved context saving and agent control features

Top Tutorials

▪ How to use Claude Code subagents for task automation

▪ Secure and scale your MCP server with OAuth and Gateways

▪ Build multi-tasking agents using complex planning and subagents using LangGraph

Top Lecture

▪ GitHub CEO latest essay: How AI will transform developers

If you're enjoying AlphaSignal please forward this email to a colleague. 

It helps us keep this content free.

./TOP_NEWS

OpenAI Releases First Open-Weight Models Since GPT-2

10,045 Likes

OpenAI releases gpt-oss-120b and gpt-oss-20b, providing free model weights and code for the first time since GPT-2. You can run these models locally and fine-tune them for your tasks.


The most notable change is that gpt-oss-120b matches the top proprietary models on major benchmarks.


Model Architecture and Requirements
OpenAI’s gpt-oss models introduce a new technical stack.

  • gpt-oss-120b has 117 billion parameters and runs on a single H100 GPU (80GB).

  • gpt-oss-20b requires at least 16GB of RAM on a consumer desktop or laptop.

  • Both models use mixture-of-experts (MoE) and native MXFP4 quantization in the MoE layer.

  • The models run with Transformers, vLLM, Ollama, LM Studio, and reference implementations.

Capabilities and Supported Tasks
You can use these models for advanced reasoning and agent workflows.

  • Chain-of-thought reasoning with configurable effort levels for different applications.

  • Built-in support for instruction following, coding, and function calling.

  • Native tool use includes web browsing and Python code execution.

  • Both models output in OpenAI’s harmony response format for compatibility with open-source tools.

Benchmark Performance
The models achieve high scores on standard AI benchmarks, closely matching OpenAI’s latest proprietary models.

  • gpt-oss-120b: 90.0% on MMLU, 2622 on Codeforces, 80.1% on GPQA Diamond.

  • gpt-oss-20b: 85.3% on MMLU, 2516 on Codeforces, 71.5% on GPQA Diamond.

  • Outperforms previous open models like DeepSeek R1 and Qwen in key areas.

Download

Deploy NVIDIA GPU Clusters in Minutes, Not Months

The ESG report shows Lambda 1-Click Clusters cut through cloud bottlenecks. Provision secure NVIDIA GPU clusters without complex manual setup or hidden charges.

  • Fast launch, clusters deploy in minutes using pre-validated hardware

  • Hardware-level isolation for security on day one

  • One workflow for compute, storage, and networking management

  • No extra charges for data ingress or egress

  • Start with full Lambda Stack: PyTorch, TensorFlow, NVIDIA drivers


Get the details in the ESG Technical Validation report.

Access the report

partner with us

./TOP_SIGNALS

OpenAI updates ChatGPT, adding digital wellbeing features like break reminders and distress detection

5,837 Likes

Hugging Face publishes Ultra-Scale Playbook, a free guide that breaks down DeepSeek, Mistral, and Meta scaling strategies

5,937 Likes

GitHub launches OpenAI-compatible inference API with no setup required for all public projects

1,139 Likes

Google introduces an open-sourced Game Arena to measure AI progress in reasoning and strategy, starting with Gemini vs Grok chess matches

1,038 Likes

Anthropic boosts Claude Code with smarter memory handling, agent control, and reading local PDFs

2,846 Likes

./TOP_TUTORIALS

How to Use Claude Code Subagents for Task Automation

1,148 Likes

Learn to configure Claude Code subagents for task-specific workflows like code review, debugging, and research. Set tools, assign models, and automate multi-agent collaboration with project-specific control.

Secure and Scale Your MCP Server with OAuth and Gateways

937 Likes

Build secure, production-grade MCP servers using OAuth 2.1, AI gateways, and dynamic secrets management. Prevent token misuse, enforce user data isolation, and implement scalable observability with OpenTelemetry.

Build Multi-Tasking Agents using Complex Planning and Subagents Using LangGraph

837 Likes

Explore how Deep Agents use LangGraph’s React loop, structured state, and built-in tools for planning, file access, and subagent control to handle complex, long-horizon tasks with modular workflows.

./TOP_READ

GitHub CEO latest essay: How AI will transform developers

The article, written by GitHub CEO Thomas Dohmke, explains how AI is transforming the software developer’s role.


"A vital shift is underway in software development, one that redefines how we build, but also who we are as developers"


It combines insights from interviews with 22 developers who have integrated AI deeply into their workflow, showing the shift from skepticism to active collaboration with AI.


Key points:

  • Shows how developers progress from AI skeptics to AI strategists who design, delegate, and verify AI-generated code.

  • Explains that success with AI means mastering new skills, including AI fluency, prompt engineering, and agent orchestration.

  • Reveals that many developers now expect 90% of code to be AI-written within a few years.

  • Argues that computer science education must evolve to prioritize collaboration with AI, not just code writing.

You should read it to understand how your role, skills, and value may shift as AI becomes standard in coding workflows.

Read more

At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 200,000 developers stay informed and ahead. We’re passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.


Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We’re based in the United States. By using our site, you agree to be governed by U.S. laws.

Looking to promote your company, product, service, or event to 200,000+ AI developers? Let's work together.

Work with us