GPT-4.1 Superform The current version of GPT-4O on a large number of tasks (development, visual analysis, etc.). Three different versions are announced: GPT – 4.1, GPT – 4.1 Mini, and GPT – 4.1 Nano.
Openai continues to deliver at the start of the spring. After unveiling three models of AI around vocal terms, the AI giant updates its LLM range. After GPT-4O and GPT-4.5, place in GPT-4.1. As usual, three new models are presented: GPT-4.1, GPT-4.1, Mini its reduced version and GPT-4.1 Nano, a very small version. Review of the abilities in presence.
GPT-4.1 will not be available in Chatgpt
Unlike its predecessors, GPT-4.1 will be exclusively available via the Openai API. Openai wants to prioritize developers and professional ecosystem rather than consumer users for this new outing. The company specifies that improvements in instructions, programming and intelligence have already been gradually integrated into the latest version of GPT-4O on Chatgpt, and that this trend will continue with the next updates. So you shouldn’t expect to see GPT-4.1 in Chatgpt. Damage.
Notable fact, Openai also announces the upcoming depreciation of GPT-4.5 Preview in the API, which will be definitively disabled in three months, on July 14, 2025. A decision justified by the similar or superior performance of the youngest on many benchmarks. One of the major advances of GPT-4.1 is the considerable extension of the context window, going from 128,000 tokens for GPT-4O to 1 million tokens for the entire GPT-4.1 range. Without aligning with Llama 4 and its ten million tokens, the 4.1 window can already process a large amount of data.
OPENAI claims precisely having led to GPT-4.1 to be more reliable in its ability to identify the relevant information on the whole context, while ignoring the unrelevant elements, regardless of their position in the entry sequence. The three versions of GPT-4.1 are also all multimodal (text, image, video), almost a standard at Openai.
The four strengths of GPT-4.1
GPT-4.1 is particularly distinguished in three key areas: code, instructions monitoring, understanding long contexts and coding latency, performance is very good: the model reaches 54.6% on Swe-Bench Verified, surpassing GPT-4O of 21.4 points and even GPT-4.5 of 26.6 points. In terms of instructions monitoring, GPT-4.1 displays an improvement of 10.5 points on multichallenge compared to GPT-4O. Interesting progress to maintain the battery on complex or long prompt. Ideal for agency AI for example.
Finally, in the multimodal field, GPT-4.1 takes up a new record on video (ability to answer multiple choice questions based on videos), with a score of 72.0%, an improvement of 6.7 points compared to GPT-4O. GPT-4.1 Mini stands out for the image analysis, often surpassing GPT-4O on MMMU (understanding of diagrams, graphics and maps) and Mathvista (visual mathematical problem solving). Even the Nano version displays remarkable performance for its size, with 80.1% on MMLU.
On latency, Openai has made significant improvements: the first response time is around fifteen seconds for a context of 128,000 tokens, and can go up to thirty seconds for a context of a million tokens with the generic version of 4.1. ,,
The most economical model of Openai
Thanks to significant improvements in the efficiency of its inference infrastructure, Openai offers GPT-4.1 at a price significantly lower than that of its predecessor. Concretely, the main model is around 26% cheaper than GPT-4O, with a price of $ 2 per million tokens as a starter and $ 8 per million tokens out. GPT-4.1 mini is positioned at $ 0.40 at a starter and $ 1.60 at the output, while GPT-4.1 Nano becomes the most economical model ever offered by Openai, only $ 0.10 at a starter and $ 0.40 out. A major competitive advantage. Finally, Openai also increases the recovery on the cache of prompt 75% (against 50% previously) for these new models. The three models of the GPT-4.1 family are available today from the official API.
| Model | Input ($/1m tokens) | Input with cache ($/1m tokens) | Output ($/1m tokens) | Max context |
|---|---|---|---|---|
| GPT-4O | $ 2.50 | $ 1.25 (50% discount) | $ 10.00 | 128k tokens |
| GPT-4.1 | $ 2.00 | $ 0.50 (75% discount) | $ 8.00 | 1m tokens |
| GPT-4.1 Mini | $ 0.40 | $ 0.10 (75% discount) | $ 1.60 | 1m tokens |
| GPT-4.1 NANO | $ 0.10 | $ 0.025 (75% discount) | $ 0.40 | 1m tokens |
A roughly efficient, efficient, and cheap model: the cocktail should fully satisfy developers.
