Editor’s update, October 19, 2026: This article covers the April 14, 2025 API launch of GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano. OpenAI now lists nano for API shutdown on October 23, 2026, and GPT-4.1 and mini for January 14, 2027. These are scheduled future dates at the time of this update. ChatGPT availability follows a separate lifecycle. Check the official deprecation schedule before choosing a model.
OpenAI launched the GPT-4.1 family in its API on April 14, 2025, emphasizing coding, instruction following, and longer-context work. The release introduced three sizes, including the first nano model in the GPT family, with different speed, capability, and cost trade-offs.
At launch, GPT-4.1 was positioned as an API-only release. That description belongs to April 2025: GPT-4.1 and GPT-4.1 mini subsequently entered ChatGPT in May 2025 and were retired from ChatGPT in February 2026. The model name alone does not establish whether it is available in a particular product today.
Coding improvements and their evaluation context
In the launch evaluation, GPT-4.1 completed 54.6% of SWE-bench Verified tasks, compared with 33.2% for the tested November 20, 2024 GPT-4o version. That is a 21.4-percentage-point difference, rather than a claim about every subsequent GPT-4o version or every coding task.
The test used 477 of the benchmark’s 500 tasks because 23 could not run on OpenAI’s infrastructure. OpenAI noted that treating the omitted tasks as failures would put GPT-4.1 at 52.1%. Prompts, tools, and the surrounding agent setup also affect results.
The original article’s discussion of more reliable code diffs and larger generated files remains relevant to the release. GPT-4.1 supported up to 32,768 output tokens, but generated patches still needed review and tests. A benchmark improvement is evidence about a specific evaluation, not proof that code is ready to deploy.
Instruction following and long context
OpenAI reported a 38.3% MultiChallenge score at launch, 10.5 percentage points above the GPT-4o comparison in that evaluation. Its appendix also published a separate score using an o3-mini grader because the company found grading problems with the default grader. Results from the two grading setups should not be mixed.
All three GPT-4.1 models supported approximately one million tokens of context. Their API references specify a 1,047,576-token context window and a June 2024 knowledge cutoff. A large context window permits more material in a request; it does not guarantee perfect recall, accurate legal analysis, or successful reasoning over every document.
OpenAI’s long-context tests examined retrieval and multi-step reasoning separately. The launch also included image-understanding evaluations. This supports the family’s historical role in document, chart, and code workflows, without implying that every family member beat every competing model on every visual task.
Why mini and nano mattered at launch
GPT-4.1 mini targeted a balance of capability, cost, and latency. OpenAI’s launch comparisons said it matched or exceeded GPT-4o on intelligence evaluations while using less time and lower cost under the measured conditions. GPT-4.1 nano targeted faster tasks such as classification and autocomplete. The original fastest-and-cheapest descriptions refer to OpenAI’s lineup at that release date.
Latency claims also need their original conditions. The announcement described nano often producing a first token in under five seconds with 128,000 input tokens. That was a time-to-first-token observation, not completion of an entire one-million-token workload in five seconds.
API pricing: a complete, dated comparison
The following are the OpenAI API text-token rates published with the April 14, 2025 launch, also displayed in the individual model references when checked on October 19, 2026. All amounts are US dollars (USD) per one million tokens, charged by usage, not a monthly ChatGPT subscription price.
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-4.1 | USD 2.00 | USD 0.50 | USD 8.00 |
| GPT-4.1 mini | USD 0.40 | USD 0.10 | USD 1.60 |
| GPT-4.1 nano | USD 0.10 | USD 0.025 | USD 0.40 |
Sources: the GPT-4.1, mini, and nano model references. Cached-input rates apply only to eligible cached tokens. The launch offered an additional 50% Batch API discount. Blended examples are omitted because they depend on an assumed mix of input, output, and cache usage; they are not a separate universal token rate.
What changed after the launch
- GPT-4.5 Preview: its API shutdown occurred on July 14, 2025. The original instruction to migrate to GPT-4.1 was part of that historical transition, not a current recommendation to start a new GPT-4.1 dependency.
- ChatGPT: GPT-4.1 and mini became available in May 2025 and were retired on February 13, 2026, according to OpenAI’s ChatGPT retirement notice.
- API nano: the schedule lists
gpt-4.1-nano/gpt-4.1-nano-2025-04-14for October 23, 2026, withgpt-5.6-lunaas the listed substitute. - API GPT-4.1 and mini: the October 14, 2026 notice schedules
gpt-4.1-2025-04-14andgpt-4.1-mini-2025-04-14for January 14, 2027. It listsgpt-6-lunaas the recommended replacement and says fine-tuned versions follow their base model’s schedule.
The 2025 release remains an important step in OpenAI’s coding and long-context development. For new work, compare the current API models, verify the precise model ID and retirement date, and evaluate any replacement on representative application tasks.
[…] described GPT-5 becoming the default for signed-in users in place of GPT-4o, o3, o4-mini, GPT-4.1, and GPT-4.5. That was a ChatGPT product transition, not a declaration that all those API models […]