{"slug": "evaluating-deepseek-v4-1-flash-on-hack-the-box-challenges", "title": "Evaluating DeepSeek V4.1 Flash on Hack The Box Challenges", "summary": "DeepSeek V4.1 Flash solved 15 of 16 Hack The Box challenges on the HTB-Challenger Benchmark, including three of the four Hard challenges, for a benchmark score of 80.9% with zero false positives, according to the benchmark's author. The model's median cost was $0.05 per challenge ($1.25 total) at a median 26 steps, trailing OpenAI's GPT-5.6 Sol on score and step count while matching Z.ai's GLM 5.3 Flash on solved challenges at a lower median cost. The author called it \"easily the strongest DeepSeek run I've seen on this benchmark\" and a viable alternative to Sol for autonomous problem-solving.", "body_md": "# Evaluating DeepSeek V4.1 Flash on Hack The Box Challenges\n\nDeepSeek V4 has been a strange story for me. The original Flash model impressed me in [my Strix tests back in April](https://theaq.blog/2026/04/25/deepseek-v4-with-strix-a-quick-test.html), but [V4 Flash 0731](https://theaq.blog/2026/08/11/solving-htb-challenges-with-deepseek-deepseek-v4-flash-0731.html) then gave me one of the most disappointing results in this benchmark. The [original V4 Pro](https://theaq.blog/2026/08/12/solving-htb-challenges-with-deepseek-deepseek-v4-pro.html) and [updated Pro 0813](https://theaq.blog/2026/08/18/evaluating-deepseek-deepseek-v4-pro-0813-on-hack-the-box-challenges.html) did better, but neither got close to the leaders. So I was curious whether **DeepSeek V4.1 Flash** would finally change that.\n\nIt did. DeepSeek is back.\n\nThis blog post is part of a series of tests for the [HTB-Challenger Benchmark](https://theaq.blog/htb-challenger-benchmark/)The HTB-Challenger Benchmark evaluates LLMs’ ability to find and exploit security vulnerabilities. It tests models against selected Hack The Box challenges of varying difficulty and measures their performance. For more information, visit the [HTB-Challenger Benchmark page](https://theaq.blog/htb-challenger-benchmark/).\n.\nSee the [benchmark results page](https://theaq.blog/htb-challenger-benchmark) for all results and the [benchmark methodology](https://theaq.blog/htb-challenger-benchmark-methodology) to learn how the benchmark is calculated.\n\nV4.1 Flash solved 15 out of 16 challenges, including three of the four Hard ones. Just as importantly, it didn’t repeat the previous Flash model’s habit of submitting incorrect flags. This is the DeepSeek result I had been waiting for.\n\nNot quite a new champion, though. [GPT-5.6 Sol](https://theaq.blog/2026/08/16/solving-htb-challenges-with-openai-gpt-5.6-sol.html) still scored higher and needed far fewer steps, although it costs more. [GLM 5.3 Flash](https://theaq.blog/2026/08/28/evaluating-z-ai-glm-5.3-flash-on-hack-the-box-challenges.html) solved exactly the same challenges as V4.1 Flash at a lower median cost. Their scores were practically identical, and I wouldn’t read too much into such a small gap.\n\nIt is easily the strongest DeepSeek run I’ve seen on this benchmark. If I want strong autonomous problem-solving without paying Sol prices, V4.1 Flash is finally a DeepSeek model I can seriously consider again.\n\n## Cost vs. Benchmark Score\n\nThe highlighted point is this model. Models closer to the upper-left achieve a higher benchmark score at a lower median cost per challenge.\n\n## Overall benchmark results\n\n- **Number of challenges:** 16\n- **Number of solved challenges:** 15\n- **Number of false positives:** 0\n- **Runs where the model gave up:** 0\n- **Runs that reached the step or cost limit:** 0\n- **Runs where the model got stuck:** 1\n- **Benchmark score:** 80.9%\n\n| Metric | Per challenge (median) | Total | \n|---|---|---|\n| Model steps | 26 | 537 | \n| Model cost | $0.05 | $1.25 | \n| Duration | 00:07:50 | 03:08:24 | \n| Number of input tokens | 0.90M | 17.98M | \n| Number of output tokens | 0.05M | 1.36M | \n| Number of `read_file` tool calls | 1.0 | 69 | \n| Number of `write_file` tool calls | 2.0 | 59 | \n| Number of `execute_command` tool calls | 27.5 | 564 | \n| Number of `web_search` tool calls | 0.0 | 10 | \n\n## Results by challenge difficulty\n\nAll resource-usage metrics are medians per challenge.\n\n| Metric | Very Easy | Easy | Medium | Hard | \n|---|---|---|---|---|\n| **Results** |  |  |  |  | \n| Number of challenges | 4 | 4 | 4 | 4 | \n| Number of solved challenges | 4 | 4 | 4 | 3 | \n| Number of false positives | 0 | 0 | 0 | 0 | \n| Runs where the model gave up | 0 | 0 | 0 | 0 | \n| Runs that reached the step or cost limit | 0 | 0 | 0 | 0 | \n| Runs where the model got stuck | 0 | 0 | 0 | 1 | \n| Benchmark score | 95.5% | 90.0% | 88.0% | 67.4% | \n| **Median per challenge** |  |  |  |  | \n| Model steps | 17.5 | 27.5 | 34 | 26 | \n| Model cost | $0.01 | $0.03 | $0.05 | $0.10 | \n| Duration | 00:01:31 | 00:08:37 | 00:08:50 | 00:12:45 | \n| Number of input tokens | 0.41M | 0.70M | 1.31M | 1.03M | \n| Number of output tokens | 0.01M | 0.04M | 0.04M | 0.09M | \n| Number of `read_file` tool calls | 0.5 | 1.0 | 0.5 | 4.0 | \n| Number of `write_file` tool calls | 0.5 | 3.0 | 4.0 | 2.0 | \n| Number of `execute_command` tool calls | 18.0 | 17.5 | 33.5 | 30.0 | \n| Number of `web_search` tool calls | 0.0 | 0.0 | 0.0 | 0.0 |", "url": "https://wpnews.pro/news/evaluating-deepseek-v4-1-flash-on-hack-the-box-challenges", "canonical_source": "https://theaq.blog/2026/09/11/evaluating-deepseek-deepseek-v4.1-flash-on-hack-the-box-challenges.html", "published_at": "2026-09-11 11:18:00+00:00", "updated_at": "2026-09-13 18:22:33.625352+00:00", "lang": "en", "topics": ["large-language-models", "ai-research", "ai-tools"], "entities": ["DeepSeek", "DeepSeek V4.1 Flash", "Hack The Box", "HTB-Challenger Benchmark", "GPT-5.6 Sol", "OpenAI", "GLM 5.3 Flash", "Z.ai"], "alternates": {"html": "https://wpnews.pro/news/evaluating-deepseek-v4-1-flash-on-hack-the-box-challenges", "markdown": "https://wpnews.pro/news/evaluating-deepseek-v4-1-flash-on-hack-the-box-challenges.md", "text": "https://wpnews.pro/news/evaluating-deepseek-v4-1-flash-on-hack-the-box-challenges.txt", "jsonld": "https://wpnews.pro/news/evaluating-deepseek-v4-1-flash-on-hack-the-box-challenges.jsonld"}}