New RepoInject/BrowserInject benchmarks show Meta's Muse Spark 1.3 rivals GPT-5.6 Sol on prompt injection
maksym_andr · x · 2026-10-27
- Expected Contributions and EXPSEC AI released two new benchmarks, RepoInject v1 and BrowserInject v1, used by Meta TBD Labs to report the prompt-injection robustness of Muse Spark 1.3.
- Results show Muse Spark 1.3's robustness is comparable to GPT-5.6 Sol.
- Contributor @Nmndsingh, who worked with the Meta team, says they are building seriously good models.
More from Models
- OpenAI Announces Three New Discoveries in Math, CS, and Physics — thesaraharminta · 2026-10-27
- Reddit Users Say the Ming Image Model Is Seriously Underrated — alley103 · 2026-10-27
- GPT-6.1 Terra fills form in 53.5s vs 3:21 for Sol in computer use side-by-side test — sandersted · 2026-10-27
- DeepMind staff reply mentions "Gemini 4 models" in apparent early slip — vivnat · 2026-10-27
- OpenAI math repo adds new Lean-verified papers, including an ETH refutation claim — danintheory · 2026-10-27
- OpenAI models produce Lean-verified existence proofs for QCD and Yang–Mills — danintheory · 2026-10-27