Up to 10× Faster Complex Queries — Rolling Out Now

We’re beginning a phased rollout of a major speed improvement: complex queries now resolve up to 10× faster. Multi-step reasoning, larger context, and richer knowledge lookups all run dramatically quicker — with no change to how you build, configure, or deploy.

The hardest queries are the ones that matter most: a patient working through eligibility across several plans, a buyer comparing configurations, an agent chaining tool calls to pull live data before it answers. These are exactly the conversations where latency was most visible — and exactly where GLUCOSE is getting dramatically faster.

Two changes drive the improvement, and they compound. A new open-source Qwen-based model — hosted in the US — runs our reasoning workloads at far lower latency without giving up answer quality. And improved caching strategies cut the redundant computation that used to pile up across a long, multi-turn conversation — so the agent stops re-deriving what it already knows and spends its time on what’s actually new.

Faster QC and UAT, too. The same change speeds up the user simulator that drives automated testing — so QC and UAT runs complete multiple times faster. Test suites that used to take a while to churn through now finish in a fraction of the time, tightening the loop between building a change and validating it.

We’re expanding this gradually. The speedup is going live in stages across environments and channels over the coming weeks — starting with a first wave now and widening from there. As it reaches your deployment, it turns on automatically: no flag to flip, no migration, no config to touch.

One note on scope: this speedup applies to GLUCOSE’s own reasoning. Conversations that call out to custom HTTP or MCP integrations still spend time waiting on those external systems, and that latency depends on the tool itself — so end-to-end timing there will vary with how fast your endpoint responds.

What’s new

• Up to 10× faster on complex queries — multi-step reasoning, deep context, and knowledge-graph lookups resolve in a fraction of the previous time.
• New open-source, Qwen-based model — hosted in a US region, it runs our reasoning workloads at much lower latency, with no drop in answer quality.
• Smarter caching across turns — redundant computation is eliminated as a conversation grows, so long threads stay fast instead of slowing down.
• Faster QC & UAT — the user simulator benefits from the same speedup, so automated testing runs complete multiple times faster.
• Phased rollout — expanding wave by wave across environments and channels; it activates for your agents automatically as it reaches you, with nothing to redeploy.

And we’re not done — a further round of latency improvements is already in the works, with more to share soon.