# Instruction Following

> Source: Benchize (https://benchize.com) — https://benchize.com/benchmarks/scale%3Ainstruction_following. When citing or reusing this content, credit "Benchize" and link to https://benchize.com/benchmarks/scale%3Ainstruction_following.

The Precise Instruction Following Prompts Dataset is composed of 1,054 instruction following prompts aimed at assessing the ability of AI models to interpret and execute detailed commands, focusing on precision and specificity. A popular method for assessing LLMs on instruction following tasks is the IFEval benchmark, which focuses on evaluating LLMs using prompts containing programmatically verifiable instructions. However, scenarios in this benchmark are limited due to the requirement of being automatically evaluable. Additionally, similar to other open source benchmarks, IFEval is prone to overfitting.

Canonical URL: https://benchize.com/benchmarks/scale%3Ainstruction_following
Last updated: Wed, 21 Oct 2026 18:52:20 GMT

Reproduction and quotation must name Benchize, link to the canonical HTML URL, and retain individual source citations.

The Precise Instruction Following Prompts Dataset is composed of 1,054 instruction following prompts aimed at assessing the ability of AI models to interpret and execute detailed commands, focusing on precision and specificity. A popular method for assessing LLMs on instruction following tasks is the IFEval benchmark, which focuses on evaluating LLMs using prompts containing programmatically verifiable instructions. However, scenarios in this benchmark are limited due to the requirement of being automatically evaluable. Additionally, similar to other open source benchmarks, IFEval is prone to overfitting.

Instruction following · Models tested: 4 · Score publishers: [scale](https://labs.scale.com/leaderboard/instruction_following)

## Model results

| # | Model | Score | ± |  |
| --- | --- | --- | --- | --- |
| 1 | [DeepSeek R1](https://benchize.com/models/deepseek-r1.md) | 87.75source score | CI 85.84–89.66 |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score` \| 87.75 \| source_score \| reported \| Higher is better \| ### Configuration Reasoning effort unknown Model configuration DeepSeek R1 sourceModelVersion Not reported thinkingEnabled Not reported ### Reported statistics Uncertainty Confidence interval: 85.84–89.66 contaminationWarning Not reported rank 2 reportedConfidenceHalfWidth 1.91 ### Source [Original source](https://labs.scale.com/leaderboard/instruction_following) Source scale Original model label DeepSeek R1 Observed 2026-10-21T16:35:41.991Z Evidence SHA-256 `2c7dbe83fd053bd9116743dda0283e16adc2d63e4ae4b62d23fc991f6772ff2f` Source locator `/rows/1` |  |  |  |  |
| 2 | [GPT-4 Turbo Preview](https://benchize.com/models/gpt-4-turbo-preview.md) | 83.19source score | CI 81.88–84.5 |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score` \| 83.19 \| source_score \| reported \| Higher is better \| ### Configuration Reasoning effort unknown Model configuration GPT-4 Turbo Preview sourceModelVersion gpt-4-0125-preview thinkingEnabled Not reported ### Reported statistics Uncertainty Confidence interval: 81.88–84.5 contaminationWarning Not reported rank 9 reportedConfidenceHalfWidth 1.31 ### Source [Original source](https://labs.scale.com/leaderboard/instruction_following) Source scale Original model label GPT-4 Turbo Preview Observed 2026-10-21T16:35:41.991Z Evidence SHA-256 `2c7dbe83fd053bd9116743dda0283e16adc2d63e4ae4b62d23fc991f6772ff2f` Source locator `/rows/8` |  |  |  |  |
| 3 | [Claude 3 Opus](https://benchize.com/models/claude-3-opus.md) | 80.12source score | CI 78.58–81.66 |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score` \| 80.12 \| source_score \| reported \| Higher is better \| ### Configuration Reasoning effort unknown Model configuration Claude 3 Opus sourceModelVersion claude-3-opus-20240229 thinkingEnabled Not reported ### Reported statistics Uncertainty Confidence interval: 78.58–81.66 contaminationWarning Not reported rank 16 reportedConfidenceHalfWidth 1.54 ### Source [Original source](https://labs.scale.com/leaderboard/instruction_following) Source scale Original model label Claude 3 Opus Observed 2026-10-21T16:35:41.991Z Evidence SHA-256 `2c7dbe83fd053bd9116743dda0283e16adc2d63e4ae4b62d23fc991f6772ff2f` Source locator `/rows/15` |  |  |  |  |
| 4 | [Mistral Large](https://benchize.com/models/mistral-large.md) | 79.89source score | CI 78.22–81.56 |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score` \| 79.89 \| source_score \| reported \| Higher is better \| ### Configuration Reasoning effort unknown Model configuration Mistral Large sourceModelVersion mistral-large-2402 thinkingEnabled Not reported ### Reported statistics Uncertainty Confidence interval: 78.22–81.56 contaminationWarning Not reported rank 17 reportedConfidenceHalfWidth 1.67 ### Source [Original source](https://labs.scale.com/leaderboard/instruction_following) Source scale Original model label Mistral Large Observed 2026-10-21T16:35:41.991Z Evidence SHA-256 `2c7dbe83fd053bd9116743dda0283e16adc2d63e4ae4b62d23fc991f6772ff2f` Source locator `/rows/16` |  |  |  |  |

---
Cite as: Instruction Following — Benchize, https://benchize.com/benchmarks/scale%3Ainstruction_following (retrieved 2026-10-21)