# Best LLMs for Roleplay in 2026

> AI models ranked for roleplay and persona work — instruction adherence, persona consistency, and creative response benchmarks.

There is no dedicated roleplay benchmark, so this reporting family ranks the measurable ingredients: instruction following (IFEval, IFBench — whether the model stays in the persona and format you set), WildBench (performance on real, messy user prompts, including creative ones), and MuSR (tracking multi-step narrative state). Human preference for prose style is worth reading alongside: Claude models held the top ratings on an independent human-preference index in the September 2026 snapshot.

Canonical page: https://benchlm.ai/best/roleplay

Last updated: October 28, 2026

## Rankings

| Rank | Model | Creator | Type | Context | Score |
|------|-------|---------|------|---------|-------|
| 1 | [Qwen3.5-27B](/models/qwen3-5-27b) | Alibaba | Open Weight | 262K | 95 |
| 2 | [Agents-A1](/models/agents-a1) | InternScience | Open Weight | 262K | 94.8 |
| 3 | [Kimi K2.5](/models/kimi-k2-5) | Moonshot AI | Open Weight | 256K | 93.9 |
| 4 | [o3-mini](/models/o3-mini) | OpenAI | Proprietary | 200K | 93.9 |
| 5 | [Qwen3.5-122B-A10B](/models/qwen3-5-122b-a10b) | Alibaba | Open Weight | 262K | 93.4 |
| 6 | [Claude 3.7 Sonnet Thinking](/models/claude-3-7-sonnet-thinking) | Anthropic | Proprietary | 200K | 93.2 |
| 7 | [GLM-5](/models/glm-5) | Z.AI | Open Weight | 200K | 92.6 |
| 8 | [o1](/models/o1) | OpenAI | Proprietary | 200K | 92.2 |
| 9 | [Flint](/models/flint-40b-a2-5b) | Recursive | Open Weight | 256K | 92.1 |
| 10 | [Llama 3.3 70B](/models/llama-3-3-70b) | Meta | Open Weight | 128K | 92.1 |
| 11 | [Qwen3.5-35B-A3B](/models/qwen3-5-35b-a3b) | Alibaba | Open Weight | 262K | 91.9 |
| 12 | [Claude 3.7 Sonnet](/models/claude-3-7-sonnet) | Anthropic | Proprietary | 200K | 90.8 |
| 13 | [Mellum2.1-12B-A2.5B-Thinking](/models/mellum2-1-12b-a2-5b-thinking) | JetBrains | Open Weight | 128K | 90.6 |
| 14 | [Kimi K2](/models/kimi-k2) | Moonshot AI | Proprietary | 128K | 89.8 |
| 15 | [Nemotron Ultra 253B](/models/nemotron-ultra-253b) | NVIDIA | Open Weight | 32K | 89.5 |
| 16 | [Qwen3 235B 2507](/models/qwen3-235b-2507) | Alibaba | Open Weight | 128K | 88.7 |
| 17 | [Llama 3.1 405B](/models/llama-3-1-405b) | Meta | Open Weight | 128K | 88.6 |
| 18 | [GPT-4.1 mini](/models/gpt-4-1-mini) | OpenAI | Proprietary | 1M | 88.5 |
| 19 | [Qwen3 235B 2507 (Reasoning)](/models/qwen3-235b-2507-reasoning) | Alibaba | Open Weight | 128K | 87.8 |
| 20 | [Llama 3.1 70B](/models/llama-3-1-70b) | Meta | Open Weight | 128K | 87.5 |
| 21 | [GPT-4.1](/models/gpt-4-1) | OpenAI | Proprietary | 1M | 87.4 |
| 22 | [dots3-note Preview](/models/dots3-note-preview) | Dots Studio | Open Weight | 512K | 87.2 |
| 23 | [Qwen3.7 Plus](/models/qwen3-7-plus) | Alibaba | Proprietary | 1M | 86.8 |
| 24 | [Qwen3.7 Max](/models/qwen3-7-max) | Alibaba | Proprietary | 1M | 86.7 |
| 25 | [DeepSeek V3](/models/deepseek-v3) | DeepSeek | Open Weight | 128K | 86.1 |
| 26 | [Claude 3.5 Haiku](/models/claude-3-5-haiku) | Anthropic | Proprietary | 200K | 85.9 |
| 27 | [LFM2.5-24B-A2B](/models/lfm2-5-24b-a2b) | LiquidAI | Open Weight | 131K | 85.7 |
| 28 | [Qwen3.6 Plus](/models/qwen3-6-plus) | Alibaba | Proprietary | 1M | 85.1 |
| 29 | [MAI-Thinking-1](/models/mai-thinking-1) | Microsoft | Proprietary | 256K | 85 |
| 30 | [Qwen3 32B (Reasoning)](/models/qwen3-32b-reasoning) | Alibaba | Open Weight | 128K | 85 |
| 31 | [Phi-4-reasoning-plus](/models/phi-4-reasoning-plus) | Microsoft | Open Weight | 32K | 84.9 |
| 32 | [LFM3-1.7B](/models/lfm3-1-7b) | LiquidAI | Open Weight | 256K | 84.8 |
| 33 | [Sarvam 105B](/models/sarvam-105b) | Sarvam | Open Weight | 128K | 84.8 |
| 34 | [Spark-X2.6-4B](/models/spark-x2-6-4b) | SparkLLM | Open Weight | 1M | 84.7 |
| 35 | [Qwen3.5 397B](/models/qwen3-5-397b) | Alibaba | Open Weight | 128K | 84.6 |
| 36 | [QwQ-32B](/models/qwq-32b) | Alibaba | Open Weight | 131K | 83.9 |
| 37 | [Phi-4-reasoning](/models/phi-4-reasoning) | Microsoft | Open Weight | 32K | 83.4 |
| 38 | [Qwen3 235B-A22B (Reasoning)](/models/qwen3-235b-a22b-reasoning) | Alibaba | Open Weight | 128K | 83.4 |
| 39 | [DeepSeek-R1](/models/deepseek-r1) | DeepSeek | Open Weight | 128K | 83.3 |
| 40 | [GPT-4.1 nano](/models/gpt-4-1-nano) | OpenAI | Proprietary | 1M | 83.2 |
| 41 | [Qwen3 235B-A22B](/models/qwen3-235b-a22b) | Alibaba | Open Weight | 128K | 83.2 |
| 42 | [Qwen3 32B](/models/qwen3-32b) | Alibaba | Open Weight | 128K | 83.2 |
| 43 | [Qwen3.8 Max](/models/qwen3-8-max) | Alibaba | Open Weight | 1M | 82.8 |
| 44 | [Ternary Bonsai 2 27B](/models/ternary-bonsai-2-27b) | Prism ML | Open Weight | 262K | 82.7 |
| 45 | [K-EXAONE 2.0](/models/k-exaone-2-0) | LG AI Research | Open Weight | 262K | 82.5 |
| 46 | [Inkling-Small](/models/inkling-small) | Thinking Machines Lab | Open Weight | 1M | 82.2 |
| 47 | [Agents-A1-4B](/models/agents-a1-4b) | InternScience | Open Weight | 262K | 81.9 |
| 48 | [Nemotron 3 Ultra](/models/nemotron-3-ultra) | NVIDIA | Open Weight | 1M | 81.7 |
| 49 | [Sage-A1.5](/models/sage-a1-5) | InternScience | Open Weight | 262K | 81.6 |
| 50 | [Qwen3.8-Omni-Flash](/models/qwen3-8-omni-flash) | Alibaba | Proprietary | 1M | 81.5 |
| 51 | [Grok 4.3](/models/grok-4-3) | xAI | Proprietary | 1M | 81.3 |
| 52 | [Qwen3.8-Flash-Next](/models/qwen3-8-flash-next) | Alibaba | Open Weight | 262K | 81.3 |
| 53 | [Celeris-1](/models/celeris-1) | Celeris | Proprietary | 128K | 80.8 |
| 54 | [Beam 1.0 501B-A23B FP8](/models/reflection-beam-fp8) | Reflection AI | Open Weight | 1M | 80.7 |
| 55 | [Llama 3.1 8B](/models/llama-3-1-8b) | Meta | Open Weight | 128K | 80.4 |
| 56 | [Solar Open 2](/models/solar-open-2) | Upstage | Open Weight | 1M | 80 |
| 57 | [Inkling](/models/inkling) | Thinking Machines Lab | Open Weight | 1M | 79.8 |
| 58 | [Beam 1.0 501B-A23B](/models/reflection-beam) | Reflection AI | Open Weight | 1M | 79.6 |
| 59 | [Qwen3.8-27B](/models/qwen3-8-27b) | Alibaba | Open Weight | 262K | 79.5 |
| 60 | [Granite 4.2 8B](/models/granite-4-2-8b) | IBM | Open Weight | 128K | 79.3 |
| 61 | [Beam 1.0 501B-A23B NVFP4](/models/reflection-beam-nvfp4) | Reflection AI | Open Weight | 1M | 79.2 |
| 62 | [Granite-4.0-H-1B](/models/granite-4-0-h-1b) | IBM | Open Weight | 128K | 78.5 |
| 63 | [Granite-4.0-1B](/models/granite-4-0-1b) | IBM | Open Weight | 128K | 77.4 |
| 64 | [Granite 4.2 30B](/models/granite-4-2-30b) | IBM | Open Weight | 128K | 77.2 |
| 65 | [LFM2.5-350M](/models/lfm2-5-350m) | LiquidAI | Open Weight | 32K | 77 |
| 66 | [Mercury 2.5](/models/mercury-2-5) | Inception | Proprietary | 260K | 77 |
| 67 | [Muse Glimmer 30B](/models/muse-glimmer-30b) | Meta | Open Weight | 131K | 77 |
| 68 | [Mercury 3](/models/mercury-3) | Inception | Proprietary | 1M | 76.8 |
| 69 | [Mellum2-12B-A2.5B-Thinking](/models/mellum2-12b-a2-5b-thinking) | JetBrains | Open Weight | 128K | 76.5 |
| 70 | [MiniCPM5-2B](/models/minicpm5-2b) | OpenBMB | Open Weight | 131K | 76.5 |
| 71 | [Gemini 3.5 Flash](/models/gemini-3-5-flash) | Google | Proprietary | 1M | 76.3 |
| 72 | [A.X K2](/models/a-x-k2) | SK Telecom | Open Weight | 262K | 75.9 |
| 73 | [Mellum2-12B-A2.5B-Instruct](/models/mellum2-12b-a2-5b-instruct) | JetBrains | Open Weight | 128K | 75.8 |
| 74 | [Ternary Bonsai 8B](/models/ternary-bonsai-8b) | Prism ML | Open Weight | 64K | 75.4 |
| 75 | [Claude Opus 4.5](/models/claude-opus-4-5) | Anthropic | Proprietary | 200K | 74.5 |
| 76 | [Ling 3.0 Flash](/models/ling-3-0-flash) | InclusionAI | Open Weight | 262K | 74.5 |
| 77 | [Granite 4.2 3B](/models/granite-4-2-3b) | IBM | Open Weight | 128K | 74.3 |
| 78 | [LFM2.5-8B-A1B](/models/lfm2-5-8b-a1b) | LiquidAI | Open Weight | 128K | 74.2 |
| 79 | [Nemotron 3 Nano Omni 30B A3B](/models/nemotron-3-nano-omni-30b-a3b) | NVIDIA | Open Weight | 256K | 74.2 |
| 80 | [Ling 3.0 Flash FP8](/models/ling-3-0-flash-fp8) | InclusionAI | Open Weight | 262K | 73.4 |
| 81 | [Nemotron 3.5 Lightning 30B A3B NVFP4](/models/nemotron-3-5-lightning-30b-a3b-nvfp4) | NVIDIA | Open Weight | 1M | 72.9 |
| 82 | [1-bit Bonsai 8B](/models/bonsai-8b) | Prism ML | Open Weight | 64K | 72.4 |
| 83 | [Nemotron 3.5 Super VL 120B A12B NVFP4](/models/nemotron-3-5-super-vl-120b-a12b-nvfp4) | NVIDIA | Open Weight | 1M | 71.5 |
| 84 | [Nemotron 3.5 Super VL 120B A12B](/models/nemotron-3-5-super-vl-120b-a12b) | NVIDIA | Open Weight | 1M | 70.9 |
| 85 | [OrcaCyber-Nano-1.0](/models/orcacyber-nano-1-0) | OrcaRouter | Open Weight | 262K | 69.6 |
| 86 | [ZAYA1-8B](/models/zaya1-8b) | Zyphra | Open Weight | 131K | 69.1 |
| 87 | [Ternary Bonsai 4B](/models/ternary-bonsai-4b) | Prism ML | Open Weight | 32K | 65.4 |
| 88 | [Ternary Bonsai 1.7B](/models/ternary-bonsai-1-7b) | Prism ML | Open Weight | 32K | 65.3 |
| 89 | [MiniCPM5-1B](/models/minicpm5-1b) | OpenBMB | Open Weight | 131K | 63.5 |
| 90 | [Hy3 Preview](/models/hy3-preview) | Tencent | Open Weight | 256K | 63.1 |
| 91 | [1-bit Bonsai 4B](/models/bonsai-4b) | Prism ML | Open Weight | 32K | 62.6 |
| 92 | [Granite-4.0-H-350M](/models/granite-4-0-h-350m) | IBM | Open Weight | 32K | 61.6 |
| 93 | [LFM2.5-VL-450M](/models/lfm2-5-vl-450m) | LiquidAI | Open Weight | 128K | 61.2 |
| 94 | [LFM2.5-2.6B](/models/lfm2-5-2-6b) | LiquidAI | Open Weight | 128K | 59.2 |
| 95 | [1-bit Bonsai 1.7B](/models/bonsai-1-7b) | Prism ML | Open Weight | 32K | 58.5 |
| 96 | [Kanana-2 3B Instruct](/models/kanana-2-3b-instruct) | Kakao | Open Weight | 32K | 57.1 |
| 97 | [Ling 2.6 Flash](/models/ling-2-6-flash) | InclusionAI | Open Weight | 262K | 57 |
| 98 | [Kanana-2 1.3B Instruct](/models/kanana-2-1-3b-instruct) | Kakao | Open Weight | 32K | 56.2 |
| 99 | [Solar Pro 3](/models/solar-pro-3) | Upstage | Proprietary | 128K | 55.8 |
| 100 | [Granite-4.0-350M](/models/granite-4-0-350m) | IBM | Open Weight | 32K | 55.4 |
| 101 | [LFM2.5-230M](/models/lfm2-5-230m) | LiquidAI | Open Weight | 32K | 55.1 |
| 102 | [LFM2.5-VL-3B](/models/lfm2-5-vl-3b) | LiquidAI | Open Weight | 32K | 54 |
| 103 | [Trinity-Large-Thinking](/models/trinity-large-thinking) | Arcee AI | Open Weight | 512K | 52.3 |
| 104 | [LLaDA2.2-mini](/models/llada2-2-mini) | InclusionAI | Open Weight | 128K | 24.9 |

## Key Takeaways

- Top model: [Qwen3.5-27B](/models/qwen3-5-27b) with a score of 95
- Best open-weight option: [Qwen3.5-27B](/models/qwen3-5-27b) at #1
- Models included: 104

## Compare the Leaders

- [Qwen3.5-27B vs Agents-A1](/compare/agents-a1-vs-qwen3-5-27b)