Category Score V3 leaderboard
LMSpeed Best Models for Instruction Following
Compare AI models for constraint adherence, structured output, novel instruction generalization, and long multi-turn instructions using every eligible benchmark in the active run.
Methodology 3.0Methodology
Current answer
No model currently meets the formal ranking rules. The table has 15 Estimated models and 77 Provisional models. They are useful signals, but they are not formal ranks.
Available leaderboard data
- Models shown
- 92
- Formally ranked models
- 0
- Benchmark columns
- 3
- Dimensions with evidence
- 2/4
How to read the benchmark bars
Each bar compares models only within the same benchmark column. Bar lengths are relative to the models shown here; they are not Category Scores and cannot be compared across benchmark columns.
| Rank | Model | LMSpeed score | Constraint following | Novel-instruction generalization | Status | Evidence | Updated | |
|---|---|---|---|---|---|---|---|---|
| IFEvalundefined models | AA-IFBenchundefined models | IFBenchundefined models | ||||||
| Estimated models — unranked15 | ||||||||
| — | Qwen3.5-27BQwen | 57.1±11.9 | 95.0 | 75.6 | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Qwen3.7 PlusQwen | 56.8±11.9 | 94.6 | 78.0 | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Qwen3.7 MaxQwen | 56.8±11.9 | 94.3 | 80.5 | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Qwen3.6 PlusQwen | 55.7±11.9 | 94.3 | 75.2 | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Qwen3.5-122B-A10BQwen | 54.2±11.9 | 93.4 | 75.7 | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Kimi K2.5MoonshotAI | 54±11.9 | 93.9 | 70.2 | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Qwen3.5 | 53.7±11.9 | 92.6 | 78.8 | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GLM-5Z.ai | 52.4±11.9 | 92.6 | 72.3 | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | O1OpenAI | 51.5±11.9 | 92.2 | 70.3 | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Qwen3.5-35B-A3BQwen | 51.5±11.9 | 91.9 | 72.5 | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Claude Opus 4.5Anthropic | 45.7±11.9 | 90.9 | 43.0 | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-4.1 MiniOpenAI | 42.4±11.9 | 88.5 | 38.3 | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-4.1OpenAI | 42.1±11.9 | 87.4 | 43.0 | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | DeepSeek V3 | 39.7±11.9 | 86.1 | 34.8 | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-4.1 NanoOpenAI | 37.1±11.9 | 83.2 | 32.0 | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| Provisional models — unranked77 | ||||||||
| — | MiniMax M3MiniMax | 57.8±16.0 | — | 82.9 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Qwen3.8 MaxQwen | 57.7±16.0 | — | — | 82.8 | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Inkling SmallThinking Machines | 57.4±16.0 | — | — | 82.2 | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Grok 4.3SpaceXAI | 57±16.0 | — | 81.3 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | MiMo-V2.5-ProXiaomi | 56.3±16.0 | — | 79.9 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | InklingThinking Machines | 56.3±16.0 | — | — | 79.8 | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Qwen3.8 27BQwen | 56.1±16.0 | — | — | 79.5 | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | O3 MiniOpenAI | 55.6±16.3 | 93.9 | — | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.2 CodexOpenAI | 55.3±16.0 | — | 77.6 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Gemini 3.1 ProGoogle | 55.1±16.0 | — | 77.1 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Muse Glimmer 30BMeta | 55.1±16.0 | — | — | 77.0 | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Qwen3.6 Max PreviewQwen | 54.9±16.0 | — | 76.6 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | DeepSeek V4 Pro 0813DeepSeek | 54.9±16.0 | — | 76.5 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GLM-5.1Z.ai | 54.8±16.0 | — | 76.3 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Gemini 3.5 FlashGoogle | 54.8±16.0 | — | 76.3 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Kimi K2.6MoonshotAI | 54.7±16.0 | — | 76.0 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.4 NanoOpenAI | 54.6±16.0 | — | 75.9 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.5OpenAI | 54.6±16.0 | — | 75.9 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | MiniMax M2.7MiniMax | 54.6±16.0 | — | 75.7 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.3 CodexOpenAI | 54.4±16.0 | — | 75.4 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.2OpenAI | 54.4±16.0 | — | 75.4 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Ling-3.0-flashInclusionai | 54.1±16.0 | — | — | 74.5 | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.4OpenAI | 53.9±16.0 | — | 73.9 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Command ACohere | 53.9±16.0 | — | 73.9 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.4 MiniOpenAI | 53.6±16.0 | — | 73.3 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GLM-5.2Z.ai | 53.6±16.0 | — | 73.3 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GLM-5 TurboZ.ai | 53.6±16.0 | — | 73.2 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5OpenAI | 53.6±16.0 | — | 73.1 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.1OpenAI | 53.5±16.0 | — | 72.9 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.6 SolOpenAI | 53.4±16.0 | — | 72.7 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | O3OpenAI | 53±16.0 | — | 71.4 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.6 TerraOpenAI | 52.9±16.0 | — | 71.2 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Gemini 3 ProGoogle | 52.6±16.0 | — | 70.4 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.1 Codex MaxOpenAI | 52.5±16.0 | — | 70.0 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.1 CodexOpenAI | 52.5±16.0 | — | 70.0 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | gpt-oss-120bOpenAI | 52.1±16.0 | — | 69.0 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | MiMo-V2-Pro | 52.1±16.0 | — | 68.8 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Mistral Medium 3.5Mistral | 52.1±16.0 | — | 68.8 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GLM-4.7Z.ai | 51.8±16.0 | — | 67.9 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Qwen3.6 27BQwen | 51.7±16.0 | — | 67.6 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Step 3.7 FlashStepFun | 51.6±16.0 | — | 67.3 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | gpt-oss-20bOpenAI | 50.8±16.0 | — | 65.1 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Qwen3.6 35B A3BQwen | 50.6±16.0 | — | 64.4 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Claude Fable 5Anthropic | 50.3±16.0 | — | 63.5 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Kimi K2.7 CodeMoonshotAI | 50.2±16.0 | — | 63.1 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Hy3 previewTencent | 50.2±16.0 | — | — | 63.1 | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Claude Opus 4.8Anthropic | 49.9±16.0 | — | 62.2 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GLM-5V TurboZ.ai | 49.6±16.0 | — | 61.1 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Claude Opus 4.7 MaxAnthropic | 48.8±16.0 | — | 58.6 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Ling 2.6 FlashinclusionAI | 48.5±16.0 | — | 57.4 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Trinity Large ThinkingArcee AI | 48.2±16.0 | — | 56.3 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Gemini 3 FlashGoogle | 47.8±16.0 | — | 55.1 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | MiMo-V2-Omni | 47.3±16.0 | — | 53.5 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | DeepSeek V3.2DeepSeek | 46±16.0 | — | 49.0 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Gemini 2.5 ProGoogle | 45.9±16.0 | — | 48.7 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Claude Sonnet 4Anthropic | 45±16.0 | — | 45.4 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Claude Opus 4.6Anthropic | 44.7±16.0 | — | 44.6 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Qwen3 MaxQwen | 44.6±16.0 | — | 44.1 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Claude Opus 4.7Anthropic | 44.5±16.0 | — | 43.6 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Llama 4 MaverickMeta | 44.3±16.0 | — | 43.0 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Kimi K2MoonshotAI | 43.8±16.0 | — | 41.5 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Claude Sonnet 4.6Anthropic | 43.7±16.0 | — | 41.2 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | MiMo-V2-Flash | 43.3±16.0 | — | 39.9 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | DeepSeek R1DeepSeek | 43.3±16.0 | — | 39.6 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Llama 4 ScoutMeta | 43.2±16.0 | — | 39.5 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Mistral Medium 3Mistral | 43.2±16.0 | — | 39.3 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Gemini 2.5 FlashGoogle | 43.1±16.0 | — | 39.0 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Llama 3.1Meta | 43.1±16.0 | — | 39.0 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | DeepSeek V3.1DeepSeek | 42.7±16.0 | — | 37.8 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GLM-4.5 AirZ.ai | 42.6±16.0 | — | 37.6 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GLM-4.6Z.ai | 42.4±16.0 | — | 36.7 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Mistral Large 3 | 42.2±16.0 | — | 36.2 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Claude 3 HaikuAnthropic | 42.2±16.0 | — | 36.1 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-4oOpenAI | 41.6±16.0 | — | 34.3 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-4o MiniOpenAI | 40.5±16.0 | — | 31.0 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Phi 4Microsoft | 37.8±16.0 | — | 23.5 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | DeepSeek R1 Distill QwenDeepSeek | 37.5±16.0 | — | 22.9 | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
What this leaderboard measures
Which AI model follows task requirements more reliably?
This leaderboard covers constraint adherence, structured output, new instruction understanding, and long multi-turn execution. A current run may cover only some of these dimensions.
Four capability dimensions
The four dimensions come from the category blueprint. Available data may cover only some of them. A dimension without evidence is not presented as a verified capability.
Constraint following
1 benchmark columns currently provide evidence for this dimension.
- IFEval
Structured output
Available data has no eligible evidence for this dimension, so it does not affect this Category Score.
Novel-instruction generalization
2 benchmark columns currently provide evidence for this dimension.
- AA-IFBench
- IFBench
Multi-turn & long instructions
Available data has no eligible evidence for this dimension, so it does not affect this Category Score.
Instruction-following tasks this page can help with
- JSON and fixed-format output that must follow field and syntax rules.
- Strict workflows with several constraints and safety rules.
- Multi-turn assistants that must remember earlier requirements in a long conversation.
How to choose a model with this leaderboard
- Step 1
Check the rating status first
Only Rated models receive a rank. Estimated and Provisional models do not have a formal position.
- Step 2
Review uncertainty and evidence
When scores are close, do not rely on rank alone. Check uncertainty, dimension coverage, and benchmark count.
- Step 3
Test the real task last
A leaderboard cannot replace your own test. Check quality, speed, price, context, and provider limits together.
Test the model with your real formats and rules. Also check retry rates, long-context stability, speed, and cost.
Rating status guide
Rated
Rated means the evidence and overlap rules are met. The model can receive a formal rank.
Estimated
Estimated means there is useful evidence, but it is not enough for a formal rank.
Provisional
Provisional means evidence is limited or dimension and benchmark-family coverage is below the estimated threshold. Use the result only as an early signal.
Benchmarks and evidence sources
Evidence source names and benchmark groups come from the currently available score data. One source may contribute several benchmarks.
BenchLM
AA-IFBench, IFBench, and IFEval
How the instruction-following model ranking is built
LMSpeed combines eligible third-party benchmarks inside four fixed capability dimensions. Rated models meet the evidence and overlap requirements for a formal rank; Estimated and Provisional models remain visible without receiving a rank.
Read the Category Score methodologyLeaderboard limits
Category Scores use the third-party benchmarks currently included by LMSpeed. Tests can use different data, prompts, and scoring rules. The result is not permanent and cannot represent every real task. Test important choices with your own data and workflow.
Frequently asked questions
Which visible model has the highest formal rank now?
There is no formal number one now. The page has 15 Estimated models and 77 Provisional models. They do not have a formal rank and should not be called the winner.
Can I compare scores across different categories?
No. Each category uses different capability dimensions and evidence. A Category Score is comparable only inside the same leaderboard. Review the matching category for each task.
Are Estimated and Provisional models still useful?
They can help you find candidates, but their evidence is not complete enough for a formal rank. Review coverage and uncertainty, then test the model on a real task.
How often does the leaderboard update?
The leaderboard updates after a new completed score run is published. The current run date and methodology version appear above. LMSpeed does not promise a fixed daily or weekly schedule.
Is the number one model always best for me?
No. Your result also depends on speed, price, context length, tool support, region, and provider limits. Use the leaderboard to narrow the field, then run your own test.
Does a high instruction-following score remove the need for validation?
No. A high score only describes the current benchmarks. Production work still needs format, content, and safety checks, plus a clear path for invalid output and retries.