Ramp's post-trained Qwen beats Opus on finance QA; 10,000-agent swarm finds 7 critical bugs
AI Engineer · youtube · 2026-11-05
At AI Engineer NY, Ramp's Alexander Shevchenko explains why an application company still runs applied research instead of waiting for the next frontier model, with three projects now in production:
- Specialist model: a post-trained Qwen 3.5 that beats Opus 4.6 on financial Q&A at Haiku-level latency, powering Ramp Sheets;
- KV cache cartridges: pre-packed context giving agents 20-30x more working memory for tasks like month-end close;
- Agent swarm pen-testing: 10,000 parallel agents pentested Ramp itself and found 7 high-severity vulnerabilities, with cheap open-weight models (Kimi K2.0 at 56.6% detection vs GPT-5.5's 52.0% per the slide) nearly matching frontier models at lower cost.
He also covers token spend management and Ramp Router; more on the Ramp Labs blog.
More from coding & agent
- POKI open-sourced: a no-jailbreak iOS agent that can fully control your iPhone — alexcovo_eth · 2026-11-05
- Foresight-SWE benchmark: coding agents solve 60.7% of issues but only 14.8% without introducing bugs — OfirPress · 2026-11-05
- User: Every Client Deck Automatically Improves HQForWork's Skill File — jacob_posel · 2026-11-05
- 10 Grok Bots to Build Your First $1K: Idea Stress Tests, Competitor Stalking & More — ChrisUniverse · 2026-11-05
- Developer says Claude and Codex nearly bankrupted him by making procrastination comfortable — oran_ge · 2026-11-05
- "We're gardening now": devs rebrand their role as agent babysitters — heyneighbor · 2026-11-05