{"slug": "frontier-cs-added-benchmark-for-gemini-4-argon-on-antigravity", "title": "Frontier CS added benchmark for Gemini 4 Argon on Antigravity", "summary": "Auditing agent traces on the FrontierCS 2 benchmark revealed that the Pi Harness terminated runs whenever it hit a fallback, such as on security tasks, in 20% of runs, according to Wenhao Chai. Rerunning Gemini 4 Argon with the Antigravity Harness moved the Google DeepMind model ahead of Astra and just behind Opus 5.5, Chai said, adding he is \"looking forward to the new ckpt.\" FrontierCS 2, a benchmark for computer science research and development, bases each task on a paper, removes the paper's contribution from released code, and asks agents to outdo the original method by the authors' chosen metrics without access to the paper or web.", "body_md": "Wenhao Chai on X: \"Gemini 4 Argon is underrated 😱.\nWhen auditing the traces, we found that Pi Harness just ends the run whenever it hits a fallback (e.g. for security task) in 20% of the runs. We rerun Argon with Antigravity Harness, and now it beats Astra and sits right behind Opus 5.5. Congrats… / X\n\nWenhao Chai on X: \"Gemini 4 Argon is underrated 😱.\nWhen auditing the traces, we found that Pi Harness just ends the run whenever it hits a fallback (e.g. for security task) in 20% of the runs. We rerun Argon with Antigravity Harness, and now it beats Astra and sits right behind Opus 5.5. Congrats to the team @GoogleDeepMind and looking forward to the new ckpt.\"\n\nGemini 4 Argon is underrated 😱.\nWhen auditing the traces, we found that Pi Harness just ends the run whenever it hits a fallback (e.g. for security task) in 20% of the runs. We rerun Argon with Antigravity Harness, and now it beats Astra and sits right behind Opus 5.5. Congrats to the team @GoogleDeepMind and looking forward to the new ckpt.\n\nFrontierCS 2 is our new benchmark for CS research & development.\nEach task is based on a paper: we delete the contribution from the released code, and ask agents to outdo the original method, as judged by the authors' chosen metrics. Agents don't have access to the paper or web.\n\nGemini 4 Argon is underrated 😱.\nWhen auditing the traces, we found that Pi Harness just ends the run whenever it hits a fallback (e.g. for security task) in 20% of the runs. We rerun Argon with Antigravity Harness, and now it beats Astra and sits right behind Opus 5.5. Congrats to the team @GoogleDeepMind and looking forward to the new ckpt.\n\nFrontierCS 2 is our new benchmark for CS research & development.\nEach task is based on a paper: we delete the contribution from the released code, and ask agents to outdo the original method, as judged by the authors' chosen metrics. Agents don't have access to the paper or web.", "url": "https://wpnews.pro/news/frontier-cs-added-benchmark-for-gemini-4-argon-on-antigravity", "canonical_source": "https://twitter.com/wenhaocha1/status/2114559544156205413", "published_at": "2026-10-27 18:24:00+00:00", "updated_at": "2026-10-27 18:41:55.014301+00:00", "lang": "en", "topics": ["ai-research", "large-language-models", "ai-agents", "artificial-intelligence"], "entities": ["Wenhao Chai", "Gemini 4 Argon", "Antigravity Harness", "Pi Harness", "FrontierCS 2", "Google DeepMind", "Astra", "Opus 5.5"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/frontier-cs-added-benchmark-for-gemini-4-argon-on-antigravity", "markdown": "https://wpnews.pro/news/frontier-cs-added-benchmark-for-gemini-4-argon-on-antigravity.md", "text": "https://wpnews.pro/news/frontier-cs-added-benchmark-for-gemini-4-argon-on-antigravity.txt", "jsonld": "https://wpnews.pro/news/frontier-cs-added-benchmark-for-gemini-4-argon-on-antigravity.jsonld"}}