Together AI launches DeepSeek V4 Flash with adjustable reasoning and cheaper agentic coding
togethercompute · x · 2026-08-04
Together AI says DeepSeek V4 Flash is now live on its platform, positioning it as a cheaper path to frontier-style agent performance.
- The model is described as stronger on coding across terminal, repository, and full-stack tasks.
- It improves tool use and automation for long-running agents.
- It exposes low, high, and max reasoning effort levels.
- Together also highlights DSpark speculative decoding for faster generation.
The accompanying benchmark graphic claims V4 Flash comes close to Opus 4.8 on selected agent benchmarks while costing 1/57th as much.
Related event: Together AI Launches DeepSeek V4 Flash(3 posts)→
More from Venture
- Paul Graham says Greptile's late-2024 revenue slump has now nearly vanished — ycombinator · 2026-08-04
- Build AI products for the core users who are willing to pay — sven_ai · 2026-08-04
- A quick survey on the economics of AI gets a two-minute recommendation — annbordetsky · 2026-08-04
- Vocabulary App 'Plume' Acquired for $8.5K at 0.9x Multiple — marclou · 2026-08-04
- Creator SaaS Tool 'ReelPilot' Acquired for $5K at 2.9x Multiple — marclou · 2026-08-04
- Agibot posts full partner roster, Moonshot AI denies Hong Kong IPO rumor, DeepSeek leads token rankings — 创业邦 · 2026-08-04