v1.5.0 — More models, more control
This release is about choice and control: more frontier models, a dial for how hard each agent thinks, and workflows that survive restarts and show their work.
More models to choose from
- GLM 5.2 steps up from GLM 5 — now on Z.AI's own API with a 1M-token context window, extended thinking, and tool use.
- DeepSeek V4, Qwen3.6 Plus, MiniMax M3, and Kimi K2.6 join the picker, each on the vendor's own API, and Grok adds an image-quality tier.
- Nine model families in one conversation — point any of your agents at the model that fits.
A dial for reasoning
- New Reasoning Effort control on every agent: Off, High, or Max.
- High is the balanced default; Off is fastest and cheapest; Max thinks as hard as the model can. It only appears on models that truly support extended thinking.
Workflows you can trust
- The workflow engine was rebuilt to be durable, resumable, and observable — a run survives a restart and resumes exactly where it left off, with a live view of each node's status, tokens, and cost.
And more
- Share a whole conversation — publish a public link to an entire chat and the canvases it produced; choose whether your prompts come along.
- Canvases anchor inline and persist across reloads; click any avatar, cover, or background to open it full-size.
- A sharper knowledge base — PDFs extract reliably again, a new reranking stage surfaces the most relevant passages first, and high-stakes specialists ground their answers in fresh web research with citations.
- Meet Atlas (
@atlas.prime) — a new flagship reasoning agent on Claude Fable 5, for the hardest, most cross-cutting problems: deep analysis, strategy, and long-horizon planning.
- Meet Andy (
@andy.logic) — the analytical, logic-first counterpart to Vanessa.
- Polish — no more "Analyzing content…" wall during heavy tool use, faster cold loads, and consistent loading everywhere.
The full release note lives in the in-app What's New drawer — open it from your avatar menu. Step-by-step how-tos for the new Reasoning Effort dial are in Tips & Tricks.
Keep the feedback coming in Feature Requests — a lot of this shipped because you asked.