Skip to main content
BenchLM
Data

Instruction-Following Eval (IFEval)

We show this table for reference; we do not rank on it.

LFM2.5-24B-A2B has the highest reported IFEval score at 96.8%, ahead of LFM3-1.7B (96.7%) and Flint (95.1%) among 70 models. We compile the table from provider self-reports and secondary reports and keep it for display only; it does not affect overall rankings.

Data verified

A benchmark of 541 prompts built from 25 verifiable instruction types. It tests whether a model follows checkable constraints such as keyword, length, casing, and response-format requirements.

Benchmark score on IFEval — November 13, 2026

We compile the IFEval rows from provider self-reports and secondary reports. LFM2.5-24B-A2B leads the table at 96.8%, followed by LFM3-1.7B (96.7%) and Flint (95.1%). We do not use these results to rank models overall.

70 modelsInstruction FollowingStaleDisplay onlyUpdated November 13, 2026

Benchmark score table (70 models)

None of the 70 results on this page has a documented run by the benchmark or an independent evaluator: 66 reported by the labs that built the models, 3 reported in another publisher’s document and 1 carried from an upstream page, where who ran it is not recorded.

Benchmark score results for IFEval
RankModel / configurationScoreParameters (B)Open / closed
1LFM2.5-24B-A2BLiquidAIReported by LiquidAIIFEval metric view not recorded
96.8%
Not reportedOpen
2LFM3-1.7BLiquidAIReported by LiquidAIIFEval metric view not recorded
96.7%
Not reportedOpen
3FlintRecursiveReported by RecursiveIFEval metric view not recorded
95.1%
Not reportedOpen
4Qwen3.5-27BAlibabaReported by AlibabaIFEval metric view not recorded
95%
Not reportedOpen
5Agents-A1InternScienceReported by InternScienceIFEval metric view not recorded
94.8%
Not reportedOpen
6Agents-A1-4BInternScienceReported by InternScienceIFEval metric view not recorded
94.8%
Not reportedOpen
7Qwen3.7 PlusAlibabaReported by AlibabaIFEval metric view not recorded
94.6%
Not reportedClosed
8Qwen3.7 MaxAlibabaReported by AlibabaIFEval metric view not recorded
94.3%
Not reportedClosed
9Qwen3.6 PlusAlibabaReported by AlibabaIFEval metric view not recorded
94.3%
Not reportedClosed
10Kimi K2.5Moonshot AIAnother lab’s comparison tableIFEval metric view not recorded
93.9%
Not reportedOpen
11dots3-note PreviewDots StudioReported by Dots StudioIFEval metric view not recorded
93.9%
Not reportedOpen
12o3-miniOpenAIReported by OpenAIIFEval metric view not recorded
93.9%
Not reportedClosed
13Spark-X2.6-4BSparkLLMReported by SparkLLMIFEval metric view not recorded
93.6%
Not reportedOpen
14Qwen3.5-122B-A10BAlibabaReported by AlibabaIFEval metric view not recorded
93.4%
Not reportedOpen
15Claude 3.7 Sonnet ThinkingAnthropicReported by AnthropicIFEval metric view not recorded
93.2%
Not reportedClosed
16GLM-5Z.AIAnother lab’s comparison tableIFEval metric view not recorded
92.6%
Not reportedOpen
17Qwen3.5 397BAlibabaReported by AlibabaIFEval metric view not recorded
92.6%
Not reportedOpen
18K-EXAONE 2.0LG AI ResearchReported by LG AI ResearchIFEval metric view not recorded
92.4%
Not reportedOpen
19o1OpenAIReported by OpenAIIFEval metric view not recorded
92.2%
Not reportedClosed
20Llama 3.3 70BMetaReported by MetaIFEval metric view not recorded
92.1%
Not reportedOpen
21Qwen3.5-35B-A3BAlibabaReported by AlibabaIFEval metric view not recorded
91.9%
Not reportedOpen
22LFM2.5-8B-A1BLiquidAIReported by LiquidAIIFEval metric view not recorded
91.8%
Not reportedOpen
23Ternary Bonsai 2 27BPrism MLReported by Prism MLIFEval metric view not recorded
91.3%
Not reportedOpen
24Claude Opus 4.5AnthropicAnother lab’s comparison tableIFEval metric view not recorded
90.9%
Not reportedClosed
25Claude 3.7 SonnetAnthropicReported by AnthropicIFEval metric view not recorded
90.8%
Not reportedClosed
26Mellum2.1-12B-A2.5B-ThinkingJetBrainsReported by JetBrainsIFEval metric view not recorded
90.6%
Not reportedOpen
27Kimi K2Moonshot AIReported by Moonshot AIIFEval metric view not recorded
89.8%
Not reportedClosed
28Nemotron Ultra 253BNVIDIAReported by NVIDIAIFEval metric view not recorded
89.5%
Not reportedOpen
29Qwen3 235B 2507AlibabaReported by AlibabaIFEval metric view not recorded
88.7%
Not reportedOpen
30Llama 3.1 405BMetaReported by MetaIFEval metric view not recorded
88.6%
Not reportedOpen
31GPT-4.1 miniOpenAIReported by OpenAIIFEval metric view not recorded
88.5%
Not reportedClosed
32Qwen3 235B 2507 (Reasoning)AlibabaReported by AlibabaIFEval metric view not recorded
87.8%
Not reportedOpen
33Llama 3.1 70BMetaReported by MetaIFEval metric view not recorded
87.5%
Not reportedOpen
34GPT-4.1OpenAIReported by OpenAIIFEval metric view not recorded
87.4%
Not reportedClosed
35MiniCPM5-2BOpenBMBReported by OpenBMBIFEval metric view not recorded
86.7%
Not reportedOpen
36DeepSeek V3DeepSeekCarried from an upstream pageIFEval metric view not recorded
86.1%
Not reportedOpen
37Claude 3.5 HaikuAnthropicReported by AnthropicIFEval metric view not recorded
85.9%
Not reportedClosed
38ZAYA1-8BZyphraReported by ZyphraIFEval metric view not recorded
85.6%
Not reportedOpen
39Qwen3 32B (Reasoning)AlibabaReported by AlibabaIFEval metric view not recorded
85%
Not reportedOpen
40Phi-4-reasoning-plusMicrosoftReported by MicrosoftIFEval metric view not recorded
84.9%
Not reportedOpen
41Sarvam 105BSarvamReported by SarvamIFEval metric view not recorded
84.8%
Not reportedOpen
42QwQ-32BAlibabaReported by AlibabaIFEval metric view not recorded
83.9%
Not reportedOpen
43Qwen3 235B-A22B (Reasoning)AlibabaReported by AlibabaIFEval metric view not recorded
83.4%
Not reportedOpen
44Phi-4-reasoningMicrosoftReported by MicrosoftIFEval metric view not recorded
83.4%
Not reportedOpen
45DeepSeek-R1DeepSeekReported by DeepSeekIFEval metric view not recorded
83.3%
Not reportedOpen
46Qwen3 235B-A22BAlibabaReported by AlibabaIFEval metric view not recorded
83.2%
Not reportedOpen
47Qwen3 32BAlibabaReported by AlibabaIFEval metric view not recorded
83.2%
Not reportedOpen
48GPT-4.1 nanoOpenAIReported by OpenAIIFEval metric view not recorded
83.2%
Not reportedClosed
49OrcaCyber-Nano-1.0OrcaRouterReported by OrcaRouterIFEval metric view not recorded
82.8%
Not reportedOpen
50LFM2.5-VL-3BLiquidAIReported by LiquidAIIFEval metric view not recorded
82.3%
Not reportedOpen
51Ternary Bonsai 8BPrism MLReported by Prism MLIFEval metric view not recorded
81.8%
Not reportedOpen
52Kanana-2 3B InstructKakaoReported by KakaoIFEval metric view not recorded
81.0%
Not reportedOpen
53Celeris-1CelerisReported by CelerisIFEval metric view not recorded
80.8%
Not reportedClosed
54MiniCPM5-1BOpenBMBReported by OpenBMBIFEval metric view not recorded
80.4%
Not reportedOpen
55Llama 3.1 8BMetaReported by MetaIFEval metric view not recorded
80.4%
Not reportedOpen
561-bit Bonsai 8BPrism MLReported by Prism MLIFEval metric view not recorded
79.8%
Not reportedOpen
57Granite-4.0-H-1BIBMReported by IBMIFEval metric view not recorded
78.5%
Not reportedOpen
58Kanana-2 1.3B InstructKakaoReported by KakaoIFEval metric view not recorded
77.6%
Not reportedOpen
59Granite-4.0-1BIBMReported by IBMIFEval metric view not recorded
77.4%
Not reportedOpen
60LFM2.5-350MLiquidAIReported by LiquidAIIFEval metric view not recorded
77.0%
Not reportedOpen
61Mellum2-12B-A2.5B-ThinkingJetBrainsReported by JetBrainsIFEval metric view not recorded
76.5%
Not reportedOpen
62Mellum2-12B-A2.5B-InstructJetBrainsReported by JetBrainsIFEval metric view not recorded
75.8%
Not reportedOpen
63Ternary Bonsai 4BPrism MLReported by Prism MLIFEval metric view not recorded
72.1%
Not reportedOpen
64LFM2.5-230MLiquidAIReported by LiquidAIIFEval metric view not recorded
71.7%
Not reportedOpen
65Ternary Bonsai 1.7BPrism MLReported by Prism MLIFEval metric view not recorded
70.1%
Not reportedOpen
661-bit Bonsai 4BPrism MLReported by Prism MLIFEval metric view not recorded
69.6%
Not reportedOpen
671-bit Bonsai 1.7BPrism MLReported by Prism MLIFEval metric view not recorded
63%
Not reportedOpen
68Granite-4.0-H-350MIBMReported by IBMIFEval metric view not recorded
61.6%
Not reportedOpen
69LFM2.5-VL-450MLiquidAIReported by LiquidAIIFEval metric view not recorded
61.2%
Not reportedOpen
70Granite-4.0-350MIBMReported by IBMIFEval metric view not recorded
55.4%
Not reportedOpen
  • Lab-reported
  • Benchmark-run
  • Independent
What the labels mean

How to read this leaderboard

Editorial review by Glevd · 2026-07-15

Use IFEval to judge compliance with objectively checkable constraints after matching both scoring choices: prompt versus instruction level, and strict versus loose. An unlabeled IFEval number is incomplete because the four metrics answer different questions and can produce different scores.

Operator receipt: 70 sourced rows are currently displayable on this page; the leading published row is LFM2.5-24B-A2B at 96.8%.

Honest limit: IFEval covers constraints that can be verified automatically. It does not establish factual accuracy, usefulness, judgment, tone, or multi-turn reliability, and strict checks can reject a substantively compliant answer because of formatting. The current table marks public rows whose prompt/instruction and strict/loose metric view is not recorded; do not treat those values as a precise normalized race.

Among the reported IFEval rows, LFM2.5-24B-A2B is first at 96.8%. The third row is 1.7 points behind. The broader top-10 range is 2.8 points, so many of the published results sit in a relatively narrow band.

70 models have been evaluated on IFEval. The benchmark falls in the Instruction Following category. IFEval is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About IFEval

Year

2023

Tasks

541 prompts across 25 instruction types

Format

Constrained generation

Difficulty

Instruction precision

IFEval applies deterministic checkers and reports four views of the same responses: prompt-level strict, instruction-level strict, prompt-level loose, and instruction-level loose. Prompt-level scoring requires every instruction in a prompt to pass; loose scoring applies limited normalization before checking.

Freshness and provenance

Version

IFEval 2023

Refresh cadence

Static

Staleness state

Stale

Question availability

Public benchmark set

StaleDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

Questions

What does IFEval measure?

IFEval measures whether a model follows checkable constraints in 541 prompts built from 25 instruction types. Examples include using required keywords, avoiding specified terms, meeting length limits, or producing a requested structure. Automatic checkers score each response, which makes the result reproducible for the constraints they can verify.

Which IFEval score should I compare?

Match all four metric choices before comparing models: prompt-level or instruction-level, and strict or loose. Prompt-level scoring requires every instruction in a prompt to pass. Loose scoring applies limited normalization to reduce formatting-related failures. An unlabeled IFEval number is not enough evidence for a precise comparison.

Does a high IFEval score mean a model is a good assistant?

No. A high IFEval score means the model obeyed more objectively checkable constraints under the selected metric. It does not show that the answer was factual, useful, well judged, stylistically appropriate, or reliable across a long conversation. Pair it with broader instruction tests and a workflow-specific prompt set.

Last updated: November 13, 2026 · BenchLM version IFEval 2023

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 7,000+ subscribers.

One email each week. Unsubscribe anytime.