gemini-4POST

Gemini 4 Argon claims 77.9% on DeepSWE but reportedly still struggles with coding

Google says Gemini 4 Argon scores 77.9% on DeepSWE and supports 1M output tokens, with internal use for C++ to Rust work, but Bloomberg sources say it still struggles on some coding tasks.

Sad Creator

@SadCreatorTalksArchived 25 days agoOriginally posted 2026.10.04

Sad Creator @SadCreatorTalksOriginally posted 2026.10.04

Open original on X

Gemini 4 Argon: Google claims 77.9% on DeepSWE (above Opus 5.5 and GPT-6 Astra) plus 1M output tokens. Internal agents already doing C++ to Rust, even Fuchsia kernel work. Bloomberg: people with access say it still struggles on some coding tasks. I believe both. That's the problem.

x.com/sadcreatortalks/status/2106768944774500367Archived 2026.10.05 09:58