Tencent Hy4 flash open-weights test: full data-to-deck workflow at $0.07 per 1M input tokens
JaynitMakwana · x · 2026-11-19
The author benchmarked Tencent's open-weight Hy4 flash against DeepSeek-V4.1-Flash-1017 inside WorkBuddy on the same messy business-analysis task: a spreadsheet with monthly revenue, ad spend, acquisition, conversion rates, and channel performance, with the goal of finding growth drivers, flagging weak spots, and producing an executive deck.
- Hy4 flash moved efficiently from raw data to deliverable: understand the sheet → compute metrics → spot trends and anomalies → prioritize → build charts → organize findings → final presentation
- API pricing is roughly $0.070/1M input tokens, $0.281/1M output tokens, and $0.003/1M cached input tokens — meaningful savings for agent workflows with repeated reasoning, tool calls, and long tasks
- The full workflow and final deck were recorded for readers to judge
More from coding & agent
- Vibe Design Tips launches free: 999 tips to make vibe-coded products feel polished — sujingshen · 2026-11-19
- The final Claude prompt: deploy to Netlify, connect the domain, and self-test as a visitor — ifioknkem · 2026-11-19
- Prompt shared: use the GoDaddy MCP to have Claude search and buy your domain — ifioknkem · 2026-11-19
- Have Claude review your site as designer, engineer, and conversion specialist in one prompt — ifioknkem · 2026-11-19
- Dev shares dual-model agent setup: Opus 5.5 for strategy, Sol 6.2 for implementation, plus free skills — PawelHuryn · 2026-11-19
- This Claude workflow builds, hosts, and publishes websites with a real domain — ifioknkem · 2026-11-19