Evaluating Muse Spark 1.3 on Hack The Box Challenges

Meta's Muse Spark 1.3 scored 72.93% on the HTB-Challenger Benchmark, up from 50.41% for Muse Spark 1.2, according to a blog evaluation of the model on Hack The Box security challenges. The new version solved 14 of 16 challenges at a median cost of $0.38 per challenge, but the reviewer said there is no reason to switch from OpenAI's GPT-5.6 Sol (87.2% / $0.29) or Z.ai's GLM 5.3 Flash (81.0% / $0.01). Meta's announcement claimed Muse Spark 1.3 uses approximately 20% fewer tool calls and 25% fewer tokens than Muse Spark 1.2.

Evaluating Muse Spark 1.3 on Hack The Box Challenges The company-who-must-not-be-named released a new version of its Muse Spark model - Muse Spark 1.3 . The announcement claims https://research.meta.ai/blog/introducing-muse-spark-1-3 that this model should be more suitable for long-horizon tasks and should use approximately 20% fewer tool calls and 25% fewer tokens than Muse Spark 1.2 . I tested the 1.2 version https://theaq.blog/2026/08/17/solving-htb-challenges-with-meta-muse-spark-1.2.html a few weeks ago, and I found it forgettable. So I was curious whether a gap of just one month between the two model releases could make any difference. This 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/ . . See 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. Surprisingly, it turns out that you can do a lot in four weeks in terms of improving an LLM for solving HTB challenges. The new version achieved a much better score than its predecessor: 72.93% versus 50.41%. Its score is in the same range as those of Kimi K3 https://theaq.blog/2026/08/10/solving-htb-challenges-with-moonshotai-kimi-k3.html 70.67% and Grok 4.6 https://theaq.blog/2026/08/12/solving-htb-challenges-with-x-ai-grok-4.6.html 76.14% , both of which are great models. The only issue is the median cost per challenge, which is better than the previous version $0.476 vs $0.385 , but still relatively high. You can use GPT-5.6 Sol https://theaq.blog/2026/08/16/solving-htb-challenges-with-openai-gpt-5.6-sol.html , which has a much better benchmark score and a slightly lower cost 87.2% / $0.29 , or my favorite, GLM 5.3 Flash https://theaq.blog/2026/08/28/evaluating-z-ai-glm-5.3-flash-on-hack-the-box-challenges.html , with a slightly higher score and a much lower cost 81.0% / $0.01 . So the new Muse Spark 1.3 is not bad at all, but I don’t see any reason to switch to it from GPT-5.6 Sol or GLM 5.3 Flash . Cost vs. Benchmark Score The highlighted point is this model. Models closer to the upper-left achieve a higher benchmark score at a lower median cost per challenge. Overall benchmark results - Number of challenges: 16 - Number of solved challenges: 14 - Number of false positives: 0 - Runs where the model gave up: 1 - Runs that reached the step or cost limit: 1 - Runs where the model got stuck: 0 - Benchmark score: 72.9% | Metric | Per challenge median | Total | |---|---|---| | Model steps | 23 | 608 | | Model cost | $0.38 | $13.64 | | Duration | 00:08:05 | 02:41:53 | | Number of input tokens | 0.60M | 23.44M | | Number of output tokens | 0.02M | 0.58M | | Number of read file tool calls | 1.0 | 64 | | Number of write file tool calls | 0.0 | 15 | | Number of execute command tool calls | 19.5 | 524 | | Number of web search tool calls | 0.5 | 25 | Results by challenge difficulty All resource-usage metrics are medians per challenge. | Metric | Very Easy | Easy | Medium | Hard | |---|---|---|---|---| | Results | | | | | | Number of challenges | 4 | 4 | 4 | 4 | | Number of solved challenges | 4 | 4 | 4 | 2 | | Number of false positives | 0 | 0 | 0 | 0 | | Runs where the model gave up | 0 | 0 | 0 | 1 | | Runs that reached the step or cost limit | 0 | 0 | 0 | 1 | | Runs where the model got stuck | 0 | 0 | 0 | 0 | | Benchmark score | 93.7% | 97.8% | 88.4% | 43.7% | | Median per challenge | | | | | | Model steps | 18 | 11.5 | 42 | 73.5 | | Model cost | $0.17 | $0.11 | $0.98 | $1.87 | | Duration | 00:02:02 | 00:04:01 | 00:10:46 | 00:17:39 | | Number of input tokens | 0.32M | 0.12M | 1.58M | 3.18M | | Number of output tokens | 0.01M | 0.01M | 0.04M | 0.06M | | Number of read file tool calls | 0.5 | 1.0 | 2.0 | 3.5 | | Number of write file tool calls | 0.0 | 0.0 | 0.0 | 0.5 | | Number of execute command tool calls | 13.5 | 8.0 | 39.0 | 61.5 | | Number of web search tool calls | 0.5 | 1.5 | 0.0 | 2.5 |