# Best LLMs for Research in 2026

> AI models ranked for research work — hard knowledge, agentic web research, and deep search benchmarks.

Research work stresses two things at once: deep, reliable knowledge (GPQA, Humanity's Last Exam, frontier-science evaluations) and the ability to actually go find and synthesize sources (BrowseComp, DeepSearch-QA, GAIA). This reporting family blends both, weighted toward the hard-knowledge and browsing benchmarks that separate research-grade models from good chat models.

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

Last updated: September 29, 2026

## Rankings

| Rank | Model | Creator | Type | Context | Score |
|------|-------|---------|------|---------|-------|
| 1 | [Atria Dawn Preview](/models/atria-dawn-preview) | Shanghai Artificial Intelligence Laboratory | Open Weight | 256K | 93.9 |
| 2 | [GPT-6 Astra](/models/gpt-6-astra) | OpenAI | Proprietary | 1.05M | 93.8 |
| 3 | [GPT-5.6 Sol](/models/gpt-5-6-sol) | OpenAI | Proprietary | 1.05M | 93.4 |
| 4 | [GPT-5.6 Terra](/models/gpt-5-6-terra) | OpenAI | Proprietary | 1.05M | 90.2 |
| 5 | [GPT-5.6 Luna](/models/gpt-5-6-luna) | OpenAI | Proprietary | 1.05M | 87.8 |
| 6 | [MAI-Thinking-1](/models/mai-thinking-1) | Microsoft | Proprietary | 256K | 84.4 |
| 7 | [MiMo-V2-Flash](/models/mimo-v2-flash) | Xiaomi | Open Weight | 256K | 84 |
| 8 | [Kimi K3](/models/kimi-k3) | Moonshot AI | Pending | 1.05M | 82.9 |
| 9 | [Step 3.7 Flash](/models/step-3-7-flash) | StepFun | Open Weight | 256K | 82.6 |
| 10 | [Claude Mythos 5](/models/claude-mythos-5) | Anthropic | Proprietary | 1M+ | 82.2 |
| 11 | [Claude Opus 5](/models/claude-opus-5) | Anthropic | Proprietary | null | 82.1 |
| 12 | [Claude Opus 4.8](/models/claude-opus-4-8) | Anthropic | Proprietary | 1M | 81.2 |
| 13 | [GPT-5.2](/models/gpt-5-2) | OpenAI | Proprietary | 400K | 79.1 |
| 14 | [Qwen3 235B 2507](/models/qwen3-235b-2507) | Alibaba | Open Weight | 128K | 78.9 |
| 15 | [Gemma 4 12B](/models/gemma-4-12b) | Google | Open Weight | 256K | 78.4 |
| 16 | [Qwen3.5-122B-A10B](/models/qwen3-5-122b-a10b) | Alibaba | Open Weight | 262K | 76.8 |
| 17 | [GPT-5.5](/models/gpt-5-5) | OpenAI | Proprietary | 1M | 76.7 |
| 18 | [Claude Opus 4.6](/models/claude-opus-4-6) | Anthropic | Proprietary | 1M | 76.1 |
| 19 | [Claude Opus 4.7 (Adaptive)](/models/claude-opus-4-7-adaptive) | Anthropic | Proprietary | 1M | 76.1 |
| 20 | [Kimi K2.5 (Reasoning)](/models/kimi-k2-5-reasoning) | Moonshot AI | Proprietary | 128K | 76 |
| 21 | [GPT-5.4](/models/gpt-5-4) | OpenAI | Proprietary | 1.05M | 75.5 |
| 22 | [Qwen3.5-27B](/models/qwen3-5-27b) | Alibaba | Open Weight | 262K | 75.1 |
| 23 | [Ornith-1.5-397B](/models/ornith-1-5-397b) | Ornith AI | Open Weight | 262K | 74.7 |
| 24 | [Qwen3.5-35B-A3B](/models/qwen3-5-35b-a3b) | Alibaba | Open Weight | 262K | 74.4 |
| 25 | [dots3-note Preview](/models/dots3-note-preview) | Dots Studio | Open Weight | 512K | 74 |
| 26 | [Hy4 preview](/models/hy4-preview) | Tencent | Open Weight | 1M | 73.9 |
| 27 | [Kimi K2.6](/models/kimi-2-6) | Moonshot AI | Open Weight | 256K | 73.7 |
| 28 | [DeepSeek V4 Pro 0813](/models/deepseek-v4-pro-0813) | DeepSeek | Open Weight | 1M | 73.6 |
| 29 | [GPT-5.5 Pro](/models/gpt-5-5-pro) | OpenAI | Proprietary | 1M | 73.6 |
| 30 | [Nemotron 3 Nano Omni 30B A3B](/models/nemotron-3-nano-omni-30b-a3b) | NVIDIA | Open Weight | 256K | 73.5 |
| 31 | [GLM-5.2](/models/glm-5-2) | Z.AI | Open Weight | 1M | 73 |
| 32 | [ZAYA1-8B](/models/zaya1-8b) | Zyphra | Open Weight | 131K | 71.8 |
| 33 | [Inkling-Small](/models/inkling-small) | Thinking Machines Lab | Open Weight | 1M | 71.6 |
| 34 | [Muse Spark 1.1](/models/muse-spark-1-1) | Meta | Proprietary | 1M | 71.2 |
| 35 | [Claude Sonnet 5](/models/claude-sonnet-5) | Anthropic | Proprietary | 1M | 71.1 |
| 36 | [Claude Sonnet 4.6](/models/claude-sonnet-4-6) | Anthropic | Proprietary | 200K | 70.8 |
| 37 | [GLM-5](/models/glm-5) | Z.AI | Open Weight | 200K | 70.7 |
| 38 | [Inkling](/models/inkling) | Thinking Machines Lab | Open Weight | 1M | 70.3 |
| 39 | [Qwen3.7 Max](/models/qwen3-7-max) | Alibaba | Proprietary | 1M | 70.1 |
| 40 | [Granite 4.2 30B](/models/granite-4-2-30b) | IBM | Open Weight | 128K | 69.2 |
| 41 | [Apodex 1.1](/models/apodex-1-1) | Apodex | Proprietary | N/A | 68.5 |
| 42 | [Qwen3.8 Max](/models/qwen3-8-max) | Alibaba | Open Weight | 1M | 68.1 |
| 43 | [Step 5 Preview](/models/step-5-preview) | StepFun | Pending | 1M | 67.6 |
| 44 | [DeepSeek V4 Flash 0731](/models/deepseek-v4-flash-0731) | DeepSeek | Open Weight | 1M | 67.5 |
| 45 | [Granite 4.2 8B](/models/granite-4-2-8b) | IBM | Open Weight | 128K | 66.6 |
| 46 | [Gemini 3.5 Flash](/models/gemini-3-5-flash) | Google | Proprietary | 1M | 66.2 |
| 47 | [Qwen3.7 Plus](/models/qwen3-7-plus) | Alibaba | Proprietary | 1M | 66.2 |
| 48 | [GPT-5.4 mini](/models/gpt-5-4-mini) | OpenAI | Proprietary | 400K | 64.8 |
| 49 | [GPT-5.4 Pro](/models/gpt-5-4-pro) | OpenAI | Proprietary | 1.05M | 64.7 |
| 50 | [Kimi K2.5](/models/kimi-k2-5) | Moonshot AI | Open Weight | 256K | 64.7 |
| 51 | [DeepSeek V4.1 Flash](/models/deepseek-v4-1-flash) | DeepSeek | Open Weight | 1M | 63.9 |
| 52 | [Qwen3.8-Flash-Next](/models/qwen3-8-flash-next) | Alibaba | Open Weight | 262K | 63.8 |
| 53 | [Qwen3.8-Omni-Flash](/models/qwen3-8-omni-flash) | Alibaba | Proprietary | 1M | 63.8 |
| 54 | [Qwen3.6 Plus](/models/qwen3-6-plus) | Alibaba | Proprietary | 1M | 63.7 |
| 55 | [Claude Opus 4.5](/models/claude-opus-4-5) | Anthropic | Proprietary | 200K | 63.3 |
| 56 | [DeepSeek V3](/models/deepseek-v3) | DeepSeek | Open Weight | 128K | 63.3 |
| 57 | [Grok 4.3](/models/grok-4-3) | xAI | Proprietary | 1M | 62.5 |
| 58 | [Qwen3.5 397B](/models/qwen3-5-397b) | Alibaba | Open Weight | 128K | 62.5 |
| 59 | [Agents-A1](/models/agents-a1) | InternScience | Open Weight | 262K | 61.6 |
| 60 | [Gemma 4 E4B](/models/gemma-4-e4b) | Google | Open Weight | 128K | 61.3 |
| 61 | [Ornith-1.5-35B-A3B](/models/ornith-1-5-35b-a3b) | Ornith AI | Open Weight | 262K | 60.8 |
| 62 | [Agents-A1-4B](/models/agents-a1-4b) | InternScience | Open Weight | 262K | 60.4 |
| 63 | [GPT-5.4 nano](/models/gpt-5-4-nano) | OpenAI | Proprietary | 400K | 60.3 |
| 64 | [GLM-5.1](/models/glm-5-1) | Z.AI | Open Weight | 203K | 60.2 |
| 65 | [Muse Spark](/models/muse-spark) | Meta | Proprietary | 262K | 60.2 |
| 66 | [Qwen3.6-27B](/models/qwen3-6-27b) | Alibaba | Open Weight | 262K | 60.2 |
| 67 | [Ling 3.0 Flash](/models/ling-3-0-flash) | InclusionAI | Open Weight | 262K | 60 |
| 68 | [Qwen3.8-27B](/models/qwen3-8-27b) | Alibaba | Open Weight | 262K | 60 |
| 69 | [ZAYA1-74B-Preview](/models/zaya1-74b-preview) | Zyphra | Open Weight | 256K | 60 |
| 70 | [Nemotron 3.5 Lightning 30B A3B NVFP4](/models/nemotron-3-5-lightning-30b-a3b-nvfp4) | NVIDIA | Open Weight | 1M | 59.8 |
| 71 | [Gemma 4 31B](/models/gemma-4-31b) | Google | Open Weight | 256K | 59.7 |
| 72 | [Solar Pro 4](/models/solar-pro-4) | Upstage | Proprietary | 512K | 58.5 |
| 73 | [Qwen3.6-35B-A3B](/models/qwen3-6-35b-a3b) | Alibaba | Open Weight | 262K | 58.2 |
| 74 | [Granite 4.2 3B](/models/granite-4-2-3b) | IBM | Open Weight | 128K | 58.1 |
| 75 | [GLM-4.7](/models/glm-4-7) | Z.AI | Open Weight | 200K | 57.2 |
| 76 | [Hy3 Preview](/models/hy3-preview) | Tencent | Open Weight | 256K | 56.4 |
| 77 | [LongCat-Flash-Lite-Sparse](/models/longcat-flash-lite-sparse) | Meituan | Open Weight | 1M | 56.3 |
| 78 | [Nemotron 3 Ultra](/models/nemotron-3-ultra) | NVIDIA | Open Weight | 1M | 56.1 |
| 79 | [Ornith-1.5-9B](/models/ornith-1-5-9b) | Ornith AI | Open Weight | 262K | 54.3 |
| 80 | [Gemini 2.5 Pro](/models/gemini-2-5-pro) | Google | Proprietary | 1M | 50.9 |
| 81 | [Gemma 4 E2B](/models/gemma-4-e2b) | Google | Open Weight | 128K | 47.6 |
| 82 | [Soofi S 30B-A3B](/models/soofi-s-30b-a3b) | Soofi Project | Open Weight | 1M | 45.4 |
| 83 | [K-EXAONE 2.0](/models/k-exaone-2-0) | LG AI Research | Open Weight | 262K | 34.6 |
| 84 | [Gemma 4 26B A4B](/models/gemma-4-26b-a4b) | Google | Open Weight | 256K | 33.6 |
| 85 | [MiniCPM5-2B](/models/minicpm5-2b) | OpenBMB | Open Weight | 131K | 24.4 |
| 86 | [LFM2.5-230M](/models/lfm2-5-230m) | LiquidAI | Open Weight | 32K | 24.1 |
| 87 | [LFM2.5-VL-450M](/models/lfm2-5-vl-450m) | LiquidAI | Open Weight | 128K | 24.1 |

## Key Takeaways

- Top model: [Atria Dawn Preview](/models/atria-dawn-preview) with a score of 93.9
- Best open-weight option: [Atria Dawn Preview](/models/atria-dawn-preview) at #1
- Models included: 87

## Compare the Leaders

- [Atria Dawn Preview vs GPT-6 Astra](/compare/atria-dawn-preview-vs-gpt-6-astra)