Internal API instruction following (OpenAI)
Models answer prompts drawn from real developer use cases that carry instructions on formatting, verbosity and length. OpenAI reports a hard subset.
OpenAI introduced it with GPT-4.1 on April 14th, 2025 and reported it again in the GPT-5 developer post on August 7th, 2025. The grading method is not described, so model-judged is assumed.
- Made by
- OpenAI
- Released
- April 2025
- Kind
- Following precise instructions
- Level
- Everyday
- Scored
- Graded by a model
- Answers
- Open-ended
- Tasks
- Kept private
- Strongest model tested
- GPT-5 mini #91 on the capability index
- Last in a launch post
- August 2025, dropped before it saturated
- Last model added
- August 2025
- Scores refreshed
- November 12th, 2026
GPT-4.5 leads with 54% among models released by April 2025
Best score per model from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best.
GPT-5 mini leads with 65.8%, claims included
Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model.
The newest model with a score here came out in April 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- Claude Fable 5.2
-
Gemini 4 Carbon - GPT-6 Astra
- Claude Opus 5.5
- and 16 more
- 1 GPT-4.5 54%
- 2 o1 high 51.3%
- 3 o3-mini high 50%
- 4 GPT-4.1 49.1%
- 5 GPT-4.1 mini 45.1%
- 6 GPT-4.1 nano 31.6%
- 7 GPT-4o Nov 2024 29.2%
- 8 GPT-4o mini 27.2%
Source: Official Internal API instruction following (OpenAI) leaderboard
- 1 GPT-5 mini 65.8%
- 2 GPT-5 nano 56.1%
- 3 GPT-4.5 54%
- 4 o1 high 51.3%
- 5 o3-mini high 50%
- 6 GPT-4.1 49.1%
- 7 GPT-4.1 mini 45.1%
- 8 GPT-4.1 nano 31.6%
- 9 GPT-4o Nov 2024 29.2%
- 10 GPT-4o mini 27.2%
Sources: Labs’ model cards and official Internal API instruction following (OpenAI) leaderboard
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-4.5 | 54% | Feb 2025 | Official leaderboard | |
| 2 | o1 | 51.3% | high | Dec 2024 | Official leaderboard |
| 3 | o3-mini | 50% | high | Jan 2025 | Official leaderboard |
| 4 | GPT-4.1 | 49.1% | Apr 2025 | Official leaderboard | |
| 5 | GPT-4.1 mini | 45.1% | Apr 2025 | Official leaderboard | |
| 6 | GPT-4.1 nano | 31.6% | Apr 2025 | Official leaderboard | |
| 7 | GPT-4o Nov 2024 | 29.2% | Nov 2024 | Official leaderboard | |
| 8 | GPT-4o mini | 27.2% | Jul 2024 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5 mini | 65.8% | Aug 2025 | Lab model card | |
| 2 | GPT-5 nano | 56.1% | Aug 2025 | Lab model card | |
| 3 | GPT-4.5 | 54% | Feb 2025 | Official leaderboard | |
| 4 | o1 | 51.3% | high | Dec 2024 | Official leaderboard |
| 5 | o3-mini | 50% | high | Jan 2025 | Official leaderboard |
| 6 | GPT-4.1 | 49.1% | Apr 2025 | Official leaderboard | |
| 7 | GPT-4.1 mini | 45.1% | Apr 2025 | Official leaderboard | |
| 8 | GPT-4.1 nano | 31.6% | Apr 2025 | Official leaderboard | |
| 9 | GPT-4o Nov 2024 | 29.2% | Nov 2024 | Official leaderboard | |
| 10 | GPT-4o mini | 27.2% | Jul 2024 | Official leaderboard |
Labs report scores for 4 models in their cards. Switch the scores above to rank them together.
Number one over time 2 models have held number one with claims included since its release in April 2025
Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.
Sources: Official Internal API instruction following (OpenAI) leaderboard and labs’ model cards
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Sources: Official Internal API instruction following (OpenAI) leaderboard and labs’ model cards
No launch post has listed it since August 2025
Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.
| Date | Launch post | Score | As the post writes it |
|---|---|---|---|
| Aug 7, 2025 | GPT-5 nano | 56.1 | Internal API instruction following eval (hard) |
| Aug 7, 2025 | GPT-5 mini | 65.8 | Internal API instruction following eval (hard) |
| Apr 14, 2025 | GPT-4.1 nano | 31.6 | Internal API instruction following (hard) |
| Apr 14, 2025 | GPT-4.1 mini | 45.1 | Internal API instruction following (hard) |
2 lab claims next to the leaderboard
Each score a lab reported for its own model, beside the same model's best score on the leaderboard. Gaps mostly come from different harnesses, effort levels and task subsets. 2 of them match, which usually means one side copied the other. Every benchmark.
| Model | Lab says | Board says | Gap |
|---|---|---|---|
| GPT-4.1 mini | 45.1% Internal API instruction following (hard) | 45.1% Leaderboard | Same |
| GPT-4.1 nano | 31.6% Internal API instruction following (hard) | 31.6% Leaderboard | Same |
More in Instructions and chat
- IFEval In use Models answer about 500 prompts containing verifiable instructions such as word limits or required keywords. Code checks each instruction for strict and loose accuracy. Gemma 4 31B IT 98.9%
- IFBench In use 58 new out-of-domain verifiable output constraints. Scored by programmatic constraint checks (strict or loose).
Nemotron-Cascade 2 30B-A3B 82.9% - MultiChallenge In use Multi-turn conversations testing instruction retention, inference memory, versioned editing and self-coherence. Graded by LLM judge with rubrics. GPT-6.1 Sol 82%
- Multi-IF In use Models follow verifiable instructions over three-turn conversations in eight languages, extending IFEval. Scored by instruction accuracy at each turn. LFM3-1.7B 89.1%
- AdvancedIF In use Models respond to over 1,600 prompts with complex, multi-turn, and system-level instructions. A judge model checks each response against expert-written rubrics.
MAI-Thinking-1 85% - IFStruct In use Models generate JSON or YAML that must follow a requested schema and phrasing constraints, without constrained decoding. A response passes only if every structural constraint holds. Gemma 4 31B IT 95.9%