Grok Build Workflows fans one prompt out to 1,024 agents
xAI's terminal agent now plans and self-verifies at scale. Also: DeepSeek's alias deadline bites, Claude voice mode gets hands, Microsoft undercuts OpenAI.
Copy markdown
One prompt, up to 1,024 agents
Grok Build's new Workflows feature takes a plain-English task, plans it, fans it across 128 parallel agents by default (1,024 for big jobs), runs independent skeptic agents to verify each finding, and hands back a single consolidated report. A built-in /deep-research workflow plus pause/resume means a whole codebase audit or issue-triage sweep can run in the background.
Save one, run it like a slash command
Workflows save into a .grok/workflows/ directory and rerun as team-shared slash commands — the exact muscle memory Claude Code and Codex users already have. It rides on the Grok Build CLI (beta since May) with full ACP support and MCP servers out of the box, so you can wire the orchestrator into your own agent apps.
The catch: parallelism is a Heavy-tier perk
The full agent budget needs SuperGrok Heavy at ~$300/month; SuperGrok and X Premium Plus get lighter access. Prefer the API? Grok Build runs about $1 per million input tokens and $2 per million output — cheap per token, but 1,024 agents add up fast, so scope your fan-out.
DeepSeek's alias deadline already bit
The deepseek-chat and deepseek-reasoner model aliases were retired on July 24, so any integration still hardcoding those names is now failing. Point calls at the explicit V4 IDs (deepseek-v4 / deepseek-v4-flash) — a two-line change that's silently broken plenty of pipelines this week.
Claude's voice mode gets hands
Claude voice now runs on Opus, Sonnet and Haiku (pick per conversation, and switch mid-call), and it can take connected actions in Gmail, Slack and other tools instead of just talking. It also picks up eight-plus languages, closing the gap with ChatGPT's desktop voice from earlier this week.
Microsoft's own models undercut OpenAI
MAI-Image-2.5-Pro and MAI-Voice-2-Flash are now live in Microsoft Foundry and rolling into Copilot and Excel. Microsoft claims the in-house models cut costs up to 89% versus comparable OpenAI options — a real lever if you build on Azure and your image or TTS bill is climbing.