Skip to main content
Back

OpenAI Reveals GPT‑4.1 Models With 1M Token Window

TRY NOW

April 16, 2025 6 min read

Hey,

Welcome to AlphaSignal – the most read newsletter by AI developers. 


We bring you the top 1% of news, papers, models, and repos, all summarized to keep you updated on the latest in AI.

IN TODAY'S SIGNAL

Read time: 6 min 05 sec

🎖️ Top News


📌  ElevenLabs

⚡️ Trending Signals

🗞️  Top Substacks

🧠  Top Lecture

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

It helps us keep this content free.

TOP NEWS

LLMs

OpenAI releases GPT-4.1 series with 1M-token context outperforming GPT 4.5 by 16.6% on SWE-bench

⇧ 31,385 Likes

What's New

OpenAI has released GPT‑4.1, GPT‑4.1 mini, and GPT‑4.1 nano via the API. All three models outperform GPT‑4o and GPT‑4.5 on key benchmarks. GPT‑4.1 reaches 54.6% on SWE-bench Verified and supports 1 million token context.


Key Features

The models include new behaviors and tool support for production use.

  • All models support up to 1 million token context window

  • GPT‑4.1 trained to follow diff format for efficient patching workflows

  • Predicted Outputs API improves full file rewrite latency

  • Improved multi-hop reasoning using Graphwalks and MRCR datasets

  • GPT‑4.1 family only available via API; not in ChatGPT UI

Coding performance improves across all evals

GPT‑4.1 leads SWE-bench, Aider polyglot, and real-world frontend code generation.

  • Scores 54.6% on SWE-bench Verified vs 38.0% (GPT‑4.5) and 33.2% (GPT‑4o)

  • Achieves 52.9% on Aider diff format, up from 18.2% with GPT‑4o

  • Makes extraneous edits in 2% of completions vs 9% in GPT‑4o

  • Preferred by human graders 80% of the time for frontend app generation

  • Supports 32K output tokens for full file rewrites

Instruction Following

The models handle structured prompts and multi-turn requests more reliably.

  • Maintains coherence and follows complex formatting across turns

  • Handles content requirements and negative instructions more accurately

  • Offers detailed step-level logs for observability and debugging

  • Scores 49% on OpenAI’s internal instruction eval (hard subset)

  • Reaches 38.3% on MultiChallenge, up from 27.8% in GPT‑4o

  • Achieves 87.4% on IFEval vs 81.0% for GPT‑4o


Long Context

All models retrieve, track, and reason over large inputs.

  • Needle-retrieval accuracy stays consistent across 1M-token inputs

  • Scores 46.3% on OpenAI-MRCR (2-needle, 1M tokens)

  • Reaches 61.7% on Graphwalks BFS and 58.0% on Graphwalks Parents

  • GPT‑4.1 mini and nano also support 1M-token contexts

  • Improves disambiguation in multi-turn, multi-needle tasks


Latency

You can choose based on speed, cost, and accuracy tradeoffs.

  • GPT‑4.1 nano returns first token in under 5 seconds at 128K input

  • GPT‑4.1 mini cuts latency by 50% vs GPT‑4o

  • Prompt caching now discounts input tokens by 75% on repeated queries

Transition from GPT‑4.5 by July 14

OpenAI will deprecate GPT‑4.5 Preview in the API in three months.

  • Replace gpt-4.5-preview with gpt-4.1 in API calls

  • Adjust prompts for improved instruction following behavior

  • Support for GPT‑4.1-only features like 1M context and diff mode

  • No additional cost for long-context beyond standard per-token rates

Availability and access
You can use GPT‑4.1 models via the OpenAI API starting today.

  • Accessible through standard OpenAI endpoints and Playground

  • GPT‑4.1 not available in ChatGPT; GPT‑4o powers ChatGPT

  • Batch API supports GPT‑4.1 models with additional 50% pricing discount

  • No local or downloadable version; hosted API use only


Community Feedback


Alex Salcido

"GPT-4.1 is very good with Cursor. It one-shot adding a web search to my local LLM assistant with long-term memory."

Get Nugget
"The 1M token context is game-changing, but the real power move? GPT-4.1-nano costs just $0.50/million tokens total ($0.10 input, $0.40 output). That's 25x cheaper than GPT-4o."

Prometheus Waluig

"That’s cool, but “API-only” is a double-edged sword. If the model’s that powerful, great at coding, instruction following, and can handle a million tokens, why gate it off from direct user access? Real-world utility includes accessibility. Not every dev wants to juggle tokens and endpoints just to explore new ideas. Sometimes you want to talk to the thing, not build scaffolding around it. Let’s not confuse “developer happiness” with broader usefulness."

Try now

Deploy Human-Like Voice Agents in Minutes

Over 500,000 voice agents have been built with ElevenLabs, handling support calls, booking appointments, and tutoring users 1-on-1. Developers are shipping real products with real results.

Pick from 5,000+ human-like voices, write a system prompt, upload your knowledge base, integrate third-party tools, and connect a phone number.

That’s it and you’re live.

Join thousands of developers building production-ready agents in days, not months.

Build voice agents

partner with us →

TRENDING SIGNALS

AI Assistants

xAI introduces Grok Studio with code execution for Python, real-time code previews, and support for Google Docs, Sheets, and Slides

⇧ 12,438 Likes

AI Tool

Anthropic announces Claude Research to deliver multi-source answers in minutes with Google integration

⇧ 5,035 Likes

LLMs

Google DeepMind presents TxGemma, open multi-step reasoning LLMs fine-tuned for therapeutic tasks across 66 benchmarks

⇧ 2,249 Likes

AI Evaluation

LM Arena unveils Search Arena, a benchmark for RAG models, 7K human votes rank Gemini 2.5 as top AI search model

⇧ 865 Likes

Embodied AI

Hugging Face acquires Pollen Robotics to lead open-source robotics, launches Reachy 2, a VR-ready open-source humanoid robot

⇧ 887 Like

TOP SUBSTACK READS

AI Industry

Google Is Winning on Every AI Front

Breaks down why Google is now ahead of OpenAI, Anthropic, and Meta across every AI front—LLMs, agents, video, cloud, hardware, and more. Covers Gemini 2.5’s benchmark dominance, Google's ecosystem advantage, and why DeepMind’s quiet momentum has turned into full-spectrum market leadership.

⇧ 208 Likes

Multi Agent Collaboration

MCP vs. A2A: Friends or Foes?

Explains why Google’s A2A protocol complements MCP, addressing cross-agent collaboration, security, and state management. Breaks down the key differences between MCP’s LLM context integration and A2A’s agent communication. Helps you understand potential future shifts in Agentic systems and their impact on multi-agent architectures.

⇧ 72 Likes

AI Agents

Your first production-ready RAG Agent

Covers how to build production-ready AI agents using LangGraph for interactive game environments. Learn to create character impersonation agents, architect an agentic RAG system, and implement multiple personas for NPCs, leveraging Groq API's LLM and advanced prompt engineering techniques for dynamic gameplay.

⇧ 34 Likes

TOP LECTURE

LLM 

How to Use o1 for High-Context, High-Precision Tasks

⇧ 3,015 Likes

This podcast breaks down how to use the o1 model effectively for software engineering, reasoning tasks, and high-context workflows.

The speakers explain how o1 differs from chat models, why prompt design matters more, and how to structure inputs and outputs for optimal results. They also cover common mistakes, model quirks, and real-world use cases.

You will learn:

  • How o1 differs from chat-first models like Claude and GPT-4

  • Why prompt briefs outperform standard prompt formats in o1

  • Techniques to inject high-volume, domain-specific context effectively

  • How to guide o1 by output goals instead of process instructions

  • Evaluation methods for assessing o1’s reasoning accuracy

  • What o1 excels at: one-shot code generation, domain-specific language use, and architecture planning

  • Limitations in style generation, multi-step app creation, and latency handling

  • UI and product design considerations when building on top of o1

Watch now

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