AgentIF: instruction following in agentic scenarios (AgentIF)
We show this table for reference; we do not rank on it.
Instruction-following benchmark built from real agentic applications, with long instructions that carry many constraints.
Benchmark score on AgentIF — October 29, 2026
We compile the AgentIF rows from provider self-reports. The table contains 2 models. Nemotron 3.5 Bolt E4B BF16 has the highest published score at 25.81%, but the available coverage and evaluation setups do not establish a market leader. We do not use these results to rank models overall.
Nemotron 3.5 Bolt E4B BF16
NVIDIA
Nemotron 3.5 Bolt E4B
NVIDIA
2 modelsInstruction FollowingCurrentDisplay onlyUpdated October 29, 2026
Benchmark score table (2 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Nemotron 3.5 Bolt E4B BF16NVIDIA | 25.81% | Not reported | Open |
| 2 | Nemotron 3.5 Bolt E4BNVIDIA | 22.95% | Not reported | Open |
How to read this leaderboard
Operator receipt: 2 sourced rows are currently displayable on this page; the highest published score among these 2 models is Nemotron 3.5 Bolt E4B BF16 at 25.81%, which does not establish a market leader.
Honest limit: Only NVIDIA-run results are stored so far, from one harness. Display-only and excluded from weighted ranking.
About AgentIF
Year
2025
Tasks
Agentic system prompts and instructions
Format
Instruction success rate
Difficulty
Complex agentic instructions
NVIDIA reports it for Nemotron 3.5 Bolt E4B (October 23, 2026), run in its NeMo Gym harness, as the instruction success rate in reasoning mode.
Freshness and provenance
Version
AgentIF 2025
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
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 AgentIF measure?
Instruction-following benchmark built from real agentic applications, with long instructions that carry many constraints.
Which model scores highest on AgentIF?
Nemotron 3.5 Bolt E4B BF16 by NVIDIA currently leads with a score of 25.81% on AgentIF.
How many models are evaluated on AgentIF?
2 AI models have published results on AgentIF in the BenchLM catalog.
Compare top models on AgentIF
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 6,100+ subscribers.
One email each week. Unsubscribe anytime.