OpenAI Releases GPT-4.1 API With One Million Token Context

What Happened

OpenAI released GPT-4.1, a new large language model available through its API, offering developers access to a context window of one million tokens. The release makes GPT-4.1 one of the largest context-window models available through a major commercial API as of April 2025.

The model is available to developers through OpenAI's standard API access tiers. OpenAI also released two smaller variants, GPT-4.1 mini and GPT-4.1 nano, positioned at lower price points for applications requiring faster or cheaper inference.

Background

Context window size determines how much text, code, or data a model can process in a single request. A one-million-token window is broadly equivalent to processing several full-length novels, large codebases, or extensive document archives in a single prompt. Prior OpenAI production models available through the API supported context windows of up to 128,000 tokens.

The expansion to one million tokens places GPT-4.1 in the same context-length range as Google's Gemini 1.5 Pro and Gemini 2.5 Pro, both of which support one-million-token contexts. Google introduced the one-million-token threshold with Gemini 1.5 Pro in early 2024, and has since extended that capability to its newer Gemini 2.5 series.

OpenAI has been iterating on the GPT-4 model family since the original GPT-4 release in March 2023. The GPT-4o series introduced multimodal capabilities, while the o3 and o4-mini reasoning models, released separately, are aimed at multi-step reasoning tasks. GPT-4.1 is positioned as an API-focused model optimised for instruction following and coding performance rather than extended reasoning chains.

Model Specifications and Pricing

OpenAI stated that GPT-4.1 shows improvements over GPT-4o on coding and instruction-following benchmarks, citing internal and third-party evaluations. The company published pricing for the model at the time of release, with GPT-4.1 mini and GPT-4.1 nano offered at reduced per-token rates compared to the flagship GPT-4.1.

The three-tier release structure follows a pattern OpenAI has used with prior model families, providing developers with options across different cost and latency requirements. GPT-4.1 nano is described as the smallest and fastest of the three variants.

OpenAI noted that GPT-4.1 is available exclusively through the API and is not being deployed in ChatGPT at launch. The company said the release is directed at developers and enterprises building applications rather than end consumers.

What It Means in Practice

A one-million-token context window allows developers to submit entire large codebases, lengthy legal documents, extended research corpora, or hours of transcribed audio within a single API call without chunking or retrieval-augmented workarounds. Use cases cited by OpenAI include software development assistance across large repositories, long-document analysis, and agentic workflows requiring extended memory within a session.

The availability of three model sizes through a single API family gives development teams the option to route tasks by complexity, using smaller models for lower-cost operations and the full GPT-4.1 for tasks requiring maximum context or capability.

For enterprise customers, the one-million-token limit reduces the need for external vector database infrastructure in some document-processing pipelines, though organisations with proprietary data retrieval requirements may continue to use retrieval-augmented generation regardless of context window size.

Market Context

The API model market has seen significant activity in early 2025. Anthropic released Claude 3.7 Sonnet with extended thinking capabilities, Meta released the open-weight Llama 4 series with multimodal support, and Google DeepMind released Gemini 2.5 Pro. OpenAI's GPT-4.1 release adds a further option to an increasingly competitive field of frontier models available through commercial APIs.

OpenAI is expected to provide additional benchmark documentation and developer guidance through its technical blog and API documentation in the days following the release.