DeepSeek V4.1 Flash beta goes live with new architecture, native multimodality, up to 507 tokens/s
智东西 · wechat · 2026-09-08
DeepSeek opened beta testing for an intermediate version of DeepSeek V4.1 Flash, built on a new architecture with native multimodal support, faster speed and lower cost. Developers can call it via model name deepseek-v4.1-flash-expires-on-0910 without changing baseurl; the beta auto-expires on Sept 10.
- Pricing matches V4 Flash: idle hours at ¥0.05/¥1.5/¥4.5 per million tokens (cache-hit input / cache-miss input / output); peak hours ¥0.1/¥3/¥9
- Rate limited to 20 concurrent requests per account
- Community tests report output speeds up to 507 tokens/s, averaging over 300 tokens/s in the pelican-on-a-bicycle SVG test
DeepSeek has shipped rapidly since July (V4 Flash GA, V4 Pro, open-source DeepSeekHarness, V4 Flash Vision Exp), suggesting a new model update cycle.
More from Models
- Kimi K3.1 Pro tops MathArena open models, closing frontier gap to about 6 months — teortaxesTex · 2026-10-26
- Ruff creator Charlie Marsh: frontier models find infinite bugs in any software — charliermarsh · 2026-10-26
- Dev testing one skill across Opus 5.5 and Sol 6.1 finds instruction following 'all over the map' — lucasmeijer · 2026-10-26
- supermemory unveils learner-1, claiming continual learning for any model — julianweisser · 2026-10-26
- Anthropic says Opus 5.5 cuts cost per task by ~40% vs Opus 5 — taufiqintech · 2026-10-26
- AI search APIs ranked: top 12 within 7 points, yet prices differ nearly 7x — FinanceYF5 · 2026-10-26