DeepSeek's model is on par with Claude 3.5 Sonnet, GPT-4o and Gemini on almost all benchmarks.

The model is the DeepSeek model in question at the time (likely DeepSeek-V3, or R1). 'On par' means scores within a few points of, or above, Claude 3.5 Sonnet, GPT-4o and a comparable Gemini model on standard benchmarks (for example MMLU, MATH, HumanEval, GPQA, SWE-bench, Arena). True: it is on par on about 80% or more of the commonly reported benchmarks. Mostly true: on par on a clear majority (about 65–80%), with a few notable gaps. Mixed: on par on roughly half. Mostly false: on par on only a minority, or only on selected benchmarks. False: it trails all three models clearly on most benchmarks.

Set before the research began.

1. E1 DeepSeek-V3 Technical Report arXiv (via ar5iv) · 19 November 2024

   “DeepSeek-V3 demonstrates competitive performance, standing on par with top-tier models such as LLaMA-3.1-405B, GPT-4o, and Claude-Sonnet 3.5, while significantly outperforming Qwen2.5 72B.”

2. E2 DeepSeek-V3 Technical Report arXiv (via ar5iv) · 19 November 2024

   “On Arena-Hard, DeepSeek-V3 achieves an impressive win rate of over 86% against the baseline GPT-4-0314, performing on par with top-tier models like Claude-Sonnet-3.5-1022.”

3. E3 Claude 3.5 Sonnet vs DeepSeek-V3: Benchmarks, Pricing & Which Is Better in 2026 LLM Stats

   “Claude 3.5 Sonnet outperforms in 4 benchmarks (GPQA, MMLU, MMLU-Pro, SWE-Bench Verified), while DeepSeek-V3 is better at 1 benchmark (DROP).”

4. E4 DeepSeek-V3 Technical Report ar5iv (arXiv mirror); authored by DeepSeek-AI · 19 November 2024

   “DeepSeek-V3 demonstrates competitive performance, standing on par with top-tier models such as LLaMA-3.1-405B, GPT-4o, and Claude-Sonnet 3.5, while significantly outperforming Qwen2.5 72B.”

5. E5 DeepSeek-V3 Technical Report ar5iv (arXiv mirror); authored by DeepSeek-AI · 19 November 2024

   “Moreover, DeepSeek-V3 excels in MMLU-Pro, a more challenging educational knowledge benchmark, where it closely trails Claude-Sonnet 3.5.”

6. E6 DeepSeek-V3 Technical Report ar5iv (arXiv mirror); authored by DeepSeek-AI · 19 November 2024

   “On the factual knowledge benchmark, SimpleQA, DeepSeek-V3 falls behind GPT-4o and Claude-Sonnet, primarily due to its design focus and resource allocation.”

7. E7 DeepSeek-V3 Technical Report ar5iv (arXiv mirror); authored by DeepSeek-AI · 19 November 2024

   “On Arena-Hard, DeepSeek-V3 achieves an impressive win rate of over 86% against the baseline GPT-4-0314, performing on par with top-tier models like Claude-Sonnet-3.5-1022.”

8. E8 DeepSeek-V3 Technical Report DeepSeek-AI (arXiv) · 19 November 2024 Own record

   “Its chat version also outperforms other open-source models and achieves performance comparable to leading closed-source models, including GPT-4o and Claude-3.5-Sonnet, on a series of standard and open-ended benchmarks.”

9. E9 DeepSeek-V3 Technical Report DeepSeek-AI (arXiv) · 19 November 2024 Own record

   “On the factual knowledge benchmark, SimpleQA, DeepSeek-V3 falls behind GPT-4o and Claude-Sonnet, primarily due to its design focus and resource allocation.”

10. E10 Details about METR's preliminary evaluation of DeepSeek-V3 METR · 12 February 2025

    “DeepSeek-V3’s performance is relatively similar across all 3 datasets and consistently comparable to Claude 3.5 Sonnet (New) and better than GPT-4o.”

11. E11 Details about METR's preliminary evaluation of DeepSeek-V3 METR · 12 February 2025

    “DeepSeek-V3’s performance on AI R&D tasks is on-par with the performance of Claude 3 Opus, very slightly better than GPT-4o, and substantially worse than the performance of Claude 3.5 Sonnet (Old), Claude 3.5 Sonnet (New), and o1.”

The evidence shows that DeepSeek-V3's developers report general parity with GPT-4o and Claude 3.5 Sonnet (E1, E9), and that METR found parity on three datasets (E11). It also shows gaps on several benchmarks (E4, E7, E12). No evidence item compares the model with Gemini or establishes what share of benchmarks are on par, so the evidence cannot place the claim in a band under the criteria.