Ox Alpha mystery model scores ~63% on full DeepSWE, on par with GPT-5.6 Sol mid
kimmonismus · x · 2026-08-23
@davis7 completed a full DeepSWE run on the mystery Ox Alpha model, ending at 63% — far below the 80% from his first subset test — and roughly on par with GPT-5.6 Sol mid.
His impressions: better "voice" than Claude or GPT, decent design taste, handles subagents and long complex work well, good code quality. Downsides: sometimes leaves dead code around (less thorough than Sol), and despite decent TPS it feels slow, especially at higher reasoning levels.
@kimmonismus adds: if it really is GLM-5.3 Flash running at Sol-mid level locally on a DGX Spark, that's a fantastic deal — a 24/7 local model with power as the only cost would be a game changer.
Related event: Ox Alpha scores ~63% on full DeepSWE; earlier 80% was subset artifact(6 posts)→
More from Models
- Unredacted Anthropic filing exposes internal data-quality debate months before Opus 3 cutoff — zacharynado · 2026-10-15
- Google says teams across the company are using Gemini 4 Argon on its hardest engineering problems — CurieuxExplorer · 2026-10-15
- Perplexity's new embedding models: index with a 9B model, search with 0.6B — antoine_chaffin · 2026-10-15
- 797GB of n-gram counts in CPU RAM let 1B models match 57% more compute — PangWeiKoh · 2026-10-15
- Leak: Moonshot's Kimi flagship 'x4' done with pre-training, K3.1 in internal testing — lyraxana · 2026-10-15
- Unsloth ships GGUF quants for Nemotron 3.5: Bolt-74B runs in 24GB RAM — danielhanchen · 2026-10-15