DeepSeek V4-Pro slows to crawl on long projects; reloading reveals it's further ahead
teortaxesTex · x · 2026-08-14
A user reports that DeepSeek V4-Pro's token streaming slows down dramatically on long projects, forcing them to kill and reload the window, only to find the model has progressed further. This raises questions about the model's long-context handling.
More from Models
- Frontier Model Guide: Opus for Design, DeepSeek for Agents — sebo_gm · 2026-08-14
- Mistral Pivoting to Compute? Now Hosting Competitor Zhipu's GLM-5.2 — tengo_harambe · 2026-08-14
- Underrated Motif 3: Matches DeepSeek V4 Flash on Benchmarks — jinnyjuice · 2026-08-14
- Google exec confirms Gemini 4 is 'most ambitious pre-training run yet' — OfficialLoganK · 2026-08-14
- Anthropic's Unreleased Opus 5 Excels at Rust Optimization but Makes Stupid Mistakes — mertdumenci · 2026-08-14
- Anthropic's Suspicious Silence Hints at Upcoming 'Claudette' Drop — beffjezos · 2026-08-14