Ampersand Claims 92.2% on SWE-bench Pro HARD-51 at 70% Lower Cost
niloofar_mire · x · 2026-11-06
- Sand announced Ampersand's "Co-Coding" multi-agent mode, trained to collaborate with humans, matching frontier coding models while being faster and cheaper with no extra coding training.
- Benchmarks: 92.2% on SWE-bench Pro v2 HARD-51, matching Fable 5.1 (70% cheaper) and Gemini 4 Argon, outperforming Astra, Muse Wonder, and Kimi K3; 77.9% on DeepSWE v1.1, on par with Gemini 4 Argon and ahead of Astra, Grok-4.7, Kimi K3, and DeepSeek-v4-pro.
- Available exclusively via its Agent API, with a $250 credit for the next week.
Related event: Ampersand's Multi-Agent Co-Coding Matches Frontier Models at 70% Lower Cost(2 posts)→
More from coding & agent
- Apify RUN panel "Agents as first class citizens" lands in SF on Nov 10 — kleffew94 · 2026-11-06
- Executor v2 launches: Zapier/Retool-like automation fully drivable by code and AI agents — threepointone · 2026-11-06
- Y Combinator-backed Reflex raises $14M Series A, its Python framework hits 29k GitHub stars — ycombinator · 2026-11-06
- One diagram maps all the agent protocols: MCP, A2A, UCP, AP2, A2UI, AG-UI — rseroter · 2026-11-06
- Sutra: open-source terminal coding agent, lowest cost per solved task — cneuralnetwork · 2026-11-06
- From Greasemonkey to now: AI lets you mod any app just by describing the change — msfeldstein · 2026-11-06