Openclaw Update 3.28: But is it Enough?

OpenClaw 3.28 lands — and it's basically a non-event

The release focuses on minor improvements, not features. Most changes touch plugin X integration and ACP bindings, with a few breaking changes for users on Quinn. The creator's call: "this was one so minor that it just doesn't really matter." If you're on Quinn, re-onboard; otherwise, just run the terminal upgrade or tell your agent to update.

Grok is now the speed play

OpenClaw's integration with Grok is the headline technical addition. Per the creator's Cloudbench data, Grok scores 19% on agentic tasks but finishes in 96 minutes — the fastest model for agentic use cases. That's the tradeoff they're optimizing for: speed over raw quality.

Claude Code is the real threat

The bigger narrative is Claude moving "very aggressively." Claude Code can now read local files, pull from Telegram, dispatch tasks, and run cron jobs. Boxmining's team is already migrating internal tasks over. The creator frames OpenClaw's push on its OpenAI compatibility layer as an attempt to "be OpenAI's kind of countermeasure" to Claude.

Claude's rate-limit cuts are forcing model swaps

Claude recently reduced limits on Pro and Max coding plans. With Opus 4.6 at 51% on Wildbench vs. GPT-5.4 at 48%, the creator argues the cost difference "is massive" and the small quality gap makes the swap obvious. The verdict on the current market: if you have money, Claude Code is the pick; if you don't, OpenClaw routing to GPT-5.4, Kimi, GLM 5.1, MiniMax, or Xiaomi MiMo is the play.

Skip Claw Hub for now

Claw Hub got multiple updates this week, but the creator is explicit: "there is probably a lot of security issues" and he won't touch it until those are fixed. Treat it as off-limits until audited.

What to actually watch

The next OpenClaw releases will be incremental bug fixes. The decisive factor is whether GPT-5.4-class and Chinese models close the gap to Opus-level performance at lower cost — that's the open question, not anything in 3.28 itself.