GPT-4.1

GPT-4.1

Quick answer: GPT-4.1 is OpenAI's April 2025 GPT-4-series update, featuring a 1M-token context window — the largest in the GPT-4 family. It scores 97.1% on the AI2 Reasoning Challenge, 9.4% on CyberGym, and 32.8% on BigCodeBench, with a SHADE-Arena score of 5.3. Priced at $2 input / $8 output per 1M tokens.

At a Glance

Where GPT-4.1 leads

• 1M-token context window — largest in the GPT-4.x series, matching Gemini 2.5 Pro
• 97.1% AI2 Reasoning Challenge — strong multi-discipline science reasoning
• 9.4% CyberGym — best in the GPT-4 family for cybersecurity tasks
• Multimodal vision support
• Lower latency than o-series models — no internal reasoning overhead

Where it lags

• 5.40% Humanity's Last Exam — below the frontier (GPT-5.6 Sol: 52.7% on ALE)
• AetherCode 10.5% — weak on competitive agentic programming
• BigCodeBench 32.8% — moderate general coding performance
• Superseded by GPT-5.x for new high-performance deployments

Best for: Long-document analysis, whole-codebase tasks, instruction-heavy pipelines that require a 1M context window at GPT-4 pricing.

What GPT-4.1 Is

GPT-4.1 is an iterative update to the GPT-4 line, released April 14, 2025 alongside GPT-4.1 mini and nano. Its headline feature is a 1M-token context window — the same as Gemini 2.5 Pro, enabling full-codebase analysis and long-document workflows without chunking.

Unlike the o-series, GPT-4.1 is a standard auto-regressive model with no internal chain-of-thought reasoning. This gives it lower latency and more predictable token costs, at the trade-off of lower accuracy on multi-step reasoning tasks. It is the appropriate GPT-4 choice when the 1M context window is a hard requirement and o-series reasoning latency is unacceptable.

GPT-4.1's 97.1% AI2 Reasoning Challenge score (covering physics, chemistry, biology, and other sciences) indicates strong factual reasoning at the GPT-4 tier. Its 9.4% CyberGym score is notably higher than o4-mini (2.5%), suggesting better capability on security analysis tasks for a non-reasoning model.

Specifications

Pricing

Prompt caching provides a 75% discount on cached input. Pricing per OpenAI pricing page.

Context Window

GPT-4.1 has a 1,000,000-token context window — roughly 750 pages of text in a single request. This matches Gemini 2.5 Pro and makes GPT-4.1 a direct competitor for long-context workloads at a similar price point.

Public Benchmark Scores

GPT-4.1 vs Alternatives

GPT-4.1's main advantage over Gemini 2.5 Pro: marginally higher AI2 RC score, better CyberGym result. Gemini 2.5 Pro has higher HLE and broader multimodal capability. vs o4-mini, GPT-4.1 offers 5× more context window and better CyberGym at ~2× the cost.

Use GPT-4.1 via API

from openai import OpenAI

client = OpenAI(api_key="YOUR_API_KEY")

response = client.chat.completions.create(
    model="gpt-4.1",
    messages=[{"role": "user", "content": "Analyse this entire codebase for architectural issues..."}],
)
print(response.choices[0].message.content)

Related Models & Benchmarks

• Compare: GPT-4o · o4-mini · Gemini 2.5 Pro · GPT-4.1 mini
• Benchmarks: AI2 Reasoning Challenge · CyberGym · BigCodeBench · SHADE-Arena

Frequently Asked Questions

Specs and scores sourced from OpenAI's official GPT-4.1 announcement (April 14, 2025) and Benchgen evaluations. Pricing cited to the OpenAI pricing page. Last updated 2026-07-24.