BenchmarkActivecategory · instruction-following
IFBench
precise instruction following with novel constraints
quality51
Updated 9 h ago · first seen 11 Sept 2026
bench_01M293SPGX8Q1PSHHMPKC9R37H
- Metric
- accuracy · %
- Direction
- Higher is better
- Results
- 450
- Leader
- grok-4-3-medium 83.33%
Score history · Trinity Large Thinking 1 row
Not enough history to chart — a single observation (56.26% on 11 Sept 2026). Rows under different configurations count separately; the list below shows each one.
- 56.26%aa_slug=trinity-large-thinking · evaluator=Artificial Analysis · index_version=4.311 Sept 2026
Leaderboard 450 current results
Select models with +, then open Compare.
| # | Model | Score | Config | Evaluated | Source | Actions |
|---|---|---|---|---|---|---|
| #401R1 Distill Llama 70BDeepSeek | 27.55% | aa_slug=deepseek-r1-distill-llama-70b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #402nvidia-nemotron-nano-9b-v2NVIDIA | 27.14% | aa_slug=nvidia-nemotron-nano-9b-v2 · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #403Molmo2-8BAllen Institute for AI | 26.94% | aa_slug=molmo2-8b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #404qwen3-1.7b-instruct-reasoningAlibaba Group | 26.87% | aa_slug=qwen3-1.7b-instruct-reasoning · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #405Ministral 3 3BMistral AI | 26.8% | aa_slug=ministral-3-3b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #406minicpm-v4-6-1-3bOpenBMB | 26.73% | aa_slug=minicpm-v4-6-1-3b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #407sarvam-30bSarvam | 26.46% | aa_slug=sarvam-30b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #408Mistral Small 3Mistral AI | 26.39% | aa_slug=mistral-small-3 · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #409lfm2-8b-a1bLiquid AI | 26.26% | aa_slug=lfm2-8b-a1b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #410Llama-3.2-3BMeta AI | 26.19% | aa_slug=llama-3-2-instruct-3b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #411granite-4-0-h-nano-1bIBM | 26.19% | aa_slug=granite-4-0-h-nano-1b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #412nvidia-nemotron-nano-12b-v2-vlNVIDIA | 25.85% | aa_slug=nvidia-nemotron-nano-12b-v2-vl · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #413apertus-70b-instructSwiss AI Initiative | 25.85% | aa_slug=apertus-70b-instruct · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #414llama-3-1-nemotron-nano-4b-reasoningNVIDIA | 25.51% | aa_slug=llama-3-1-nemotron-nano-4b-reasoning · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #415exaone-4-0-1-2bLG AI Research | 25.31% | aa_slug=exaone-4-0-1-2b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #416magistral-mediumMistral AI | 25.1% | aa_slug=magistral-medium · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #417Granite 4.0 MicroIBM | 24.76% | aa_slug=granite-4-0-micro · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #418magistral-smallMistral AI | 24.76% | aa_slug=magistral-small · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #419llama-3-instruct-8bMeta AI | 24.56% | aa_slug=llama-3-instruct-8b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #420olmo-2-7bAllen Institute for AI | 24.42% | aa_slug=olmo-2-7b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #421qwen3-14b-instructAlibaba Group | 23.95% | aa_slug=qwen3-14b-instruct · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #422phi-3-miniMicrosoft | 23.88% | aa_slug=phi-3-mini · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #423ling-mini-2-0inclusionAI | 23.61% | aa_slug=ling-mini-2-0 · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #424Phi 4Microsoft | 23.54% | aa_slug=phi-4 · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #425Qwen3-0.6BQwen | 23.33% | aa_slug=qwen3-0.6b-instruct-reasoning · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #426exaone-4-0-1-2b-reasoningLG AI Research | 22.99% | aa_slug=exaone-4-0-1-2b-reasoning · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #427DeepSeek-R1-Distill-Qwen-32BDeepSeek | 22.93% | aa_slug=deepseek-r1-distill-qwen-32b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #428Llama-3.2-1BMeta AI | 22.79% | aa_slug=llama-3-2-instruct-1b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #429granite-3-3-8b-instructIBM | 22.45% | aa_slug=granite-3-3-8b-instruct · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #430apertus-8b-instructSwiss AI Initiative | 22.38% | aa_slug=apertus-8b-instruct · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #431deepseek-r1-distill-qwen-14bDeepSeek | 22.11% | aa_slug=deepseek-r1-distill-qwen-14b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #432gemma-3n-e2bGoogle | 22.04% | aa_slug=gemma-3n-e2b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #433LFM2-1.2BLiquid AI | 21.97% | aa_slug=lfm2-1-2b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #434qwen3-0.6b-instructAlibaba Group | 21.9% | aa_slug=qwen3-0.6b-instruct · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #435qwen3-5-0-8b-non-reasoningAlibaba Group | 21.56% | aa_slug=qwen3-5-0-8b-non-reasoning · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #436Qwen3.5-0.8BQwen | 21.5% | aa_slug=qwen3-5-0-8b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #437qwen3-1.7b-instructAlibaba Group | 21.09% | aa_slug=qwen3-1.7b-instruct · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #438phi-4-miniMicrosoft | 21.09% | aa_slug=phi-4-mini · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #439granite-4-0-nano-1bIBM | 20.54% | aa_slug=granite-4-0-nano-1b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #440tiny-aya-globalCohere | 20.14% | aa_slug=tiny-aya-global · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #441gemma-3-1bGoogle | 19.93% | aa_slug=gemma-3-1b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #442Mistral 7BMistral AI | 19.93% | aa_slug=mistral-7b-instruct · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #443DeepSeek-R1-0528-Qwen3-8BDeepSeek | 19.86% | aa_slug=deepseek-r1-qwen3-8b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #444molmo-7b-dAllen Institute for AI | 19.66% | aa_slug=molmo-7b-d · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #445lfm2-2-6bLiquid AI | 19.52% | aa_slug=lfm2-2-6b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #446granite-4-0-h-350mIBM | 17.55% | aa_slug=granite-4-0-h-350m · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #447deepseek-r1-distill-llama-8bDeepSeek | 17.55% | aa_slug=deepseek-r1-distill-llama-8b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #448granite-4-0-350mIBM | 15.92% | aa_slug=granite-4-0-350m · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #449DeepSeek-R1-Distill-Qwen-1.5BDeepSeek | 13.2% | aa_slug=deepseek-r1-distill-qwen-1-5b · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History | |
| #450gemma-3-270mGoogle | 12.11% | aa_slug=gemma-3-270m · evaluator=Artificial Analysis · index_version=4.3 | — (obs. 11 Sept 2026) | artificialanalysis.aiT2 | History |
Scores are reported as published, with their evaluation configuration (harness, prompting, judge). The bar is relative to the best score on this page. Results with different configs are not directly comparable — see methodology.
Definition
- Category
- instruction-following
Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 9 h agomedium
- Task
- precise instruction following with novel constraints
Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 9 h agomedium
- Metric
- accuracy · %
Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 9 h agomedium
- Direction
- Higher is better
- Paper
- https://arxiv.org/abs/2507.02833
Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 9 h agomedium
Each value shows its source, tier and observation time. Missing rows mean no source stated them. How results are recorded →
- Evaluated models
- grok-3-mini-reasoning, gemma-4-12B, grok-4-3-low, GPT-5.1-Codex Mini, Qwen3 Coder Next, KAT-Coder-Pro V2, nova-2-0-lite-reasoning-medium, lfm2-24b-a2b +442(450 total)
As of
Rewind the record: see this entity's attributes exactly as AI Atlas knew them on a given day.
Claim history
5 claims · 5 properties
Categorycategory1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| instruction-following | → current | current | AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2 | medium | curated |
Metricmetric1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| accuracy | → current | current | AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2 | medium | curated |
Paperpaper1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| https://arxiv.org/abs/2507.02833 | → current | current | AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2 | medium | curated |
Tasktask1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| precise instruction following with novel constraints | → current | current | AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2 | medium | curated |
Unitunit1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| % | → current | current | AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2 | medium | curated |
Claims are temporal and append-only: a new observation closes the previous claim (valid_to) instead of overwriting it. Conflicting claims from different sources are kept side by side and flagged — never averaged. Methodology →
| Source | Document | Type | Tier | Last observed | Snapshots |
|---|---|---|---|---|---|
| Artificial Analysis | artificialanalysis.ai/leaderboards/models | leaderboard | T2· Quality secondary | 8 h ago | 1 |
Tier 1 = official/primary, 2 = quality secondary, 3 = community, 4 = unverified. Every snapshot is archived; see all sources and the methodology.