OpenAI Launches GPT-4.1 API With Expanded Context Window

OpenAI released GPT-4.1, a new large language model available through its API, on April 14, 2025. The model supports a context window of one million tokens, the largest the company has made available in a production API endpoint, and is priced below GPT-4o.

What Happened

OpenAI made GPT-4.1 available to developers through its API alongside two smaller variants: GPT-4.1 Mini and GPT-4.1 Nano. The company said all three models are optimized for instruction following and coding tasks. GPT-4.1 is not available in ChatGPT and is positioned as a developer-facing release rather than a consumer product update.

The flagship GPT-4.1 model is priced at $2.00 per million input tokens and $8.00 per million output tokens. GPT-4.1 Mini is priced at $0.40 per million input tokens and $1.60 per million output tokens. GPT-4.1 Nano, the smallest and least expensive variant, is priced at $0.10 per million input tokens and $0.40 per million output tokens.

Background

OpenAI is the San Francisco-based company behind the ChatGPT consumer product and the GPT series of large language models. The company has held an API business serving enterprise and developer customers since 2020. GPT-4o, the prior generation flagship, was released in May 2024 and has served as the primary model in both the API and ChatGPT since that time.

The one-million-token context window in GPT-4.1 represents a significant increase from GPT-4o, which supported a maximum context of 128,000 tokens. A one-million-token context allows a model to process and respond to inputs equivalent to several full-length novels or large codebases within a single prompt.

The release comes as competition in the developer API market has intensified. Google DeepMind's Gemini 2.5 Pro, released earlier in 2025, also supports a one-million-token context window. Anthropic's Claude 3.7 Sonnet supports up to 200,000 tokens.

Performance and Benchmarks

OpenAI reported that GPT-4.1 outperforms GPT-4o on several coding and instruction-following benchmarks. On SWE-bench Verified, a benchmark measuring a model's ability to resolve real-world software engineering issues, OpenAI said GPT-4.1 scored 54.6 percent, compared to 33.2 percent for GPT-4o. The company said GPT-4.1 demonstrated improved ability to follow multi-step instructions and to avoid adding unrequested content in responses.

OpenAI also stated that GPT-4.1 shows improvements in long-context retrieval tasks, a category of evaluation that tests whether models can accurately locate and use information placed at various positions within a long input.

What It Means in Practice

The one-million-token context window allows developers building on the API to submit large documents, codebases, or conversation histories in a single API call without chunking or retrieval-augmented generation workarounds. OpenAI said the model is designed for use cases including agentic coding, document analysis, and customer support automation.

The pricing structure places GPT-4.1 below GPT-4o at current list rates, which OpenAI said reflects efficiency improvements in the model's architecture. GPT-4.1 Nano, at $0.10 per million input tokens, is OpenAI's least expensive model to date on a per-token basis.

OpenAI said GPT-4o will remain available through the API and that GPT-4.1 does not replace it immediately. The company did not announce a timeline for deprecating GPT-4o.

OpenAI said it plans to evaluate bringing GPT-4.1 capabilities to ChatGPT in future updates, though no release date for a consumer-facing rollout has been announced.