For Agent Memory, the Boring DeepSeek Non-Thinking Pass Wins on Speed and Accuracy
Brave_Pressure_9886 · reddit · 2026-08-28
A user building agent memory systems reports that on a narrow task — roughly 1,000 cached prompt tokens plus 2,000 tokens of memory material for extraction and summarization — the non-thinking pass of DeepSeek V4 Flash 0731 is unexpectedly excellent: fast and sharp, producing better summaries than Luna's low/medium runs and Terra on low effort.
Other notes: Qwen 3.5 Flash is the other non-thinking model he likes, especially for Chinese; he hasn't tested Qwen 3.7/3.8. His own private coding radar gave the V4 Flash/Pro non-thinking route 50 points vs. 8 for Luna low. He routes everything through ZenMux as a single API gateway so models can be swapped without re-integration. His core point: memory analysis sits inside a repeated workflow, so speed itself is a hard requirement — which is why DeepSeek stays in his rotation.
More from Models
- GLM-5.3 unprompted reasons about how its actions affect its future training, worries Apollo — MariusHobbhahn · 2026-10-15
- Gemini 4 argon vs Opus 5.5 hands-on: 3 min for $3 vs 25 min for $8 — zacharynado · 2026-10-15
- SentinelOne says OpenAI did 4-6 weeks of reverse-engineering in 8 hours — OpenAI · 2026-10-15
- Qwen3.8 27B scores 23.57% on word-based addition without reasoning, 167/169 with it — dl_weekly · 2026-10-15
- Hacker turns EmbeddingGemma 2 into 288x realtime STT and a fast TTS — kastnerkyle · 2026-10-15
- Claude Fable 5.5 reportedly in partners' hands, release eyed next week — lyraxana · 2026-10-15