OpenClaw 2026.2.6 Update: New Models, Security Scanner, and Cron Fixes

OpenClaw 2026.2.6 Update: New Models, Security Scanner, and Cron Fixes

OpenClaw shipped v2026.2.6 with a mix of practical upgrades (new model support, UX improvements) and hardening (a code safety scanner, credential redaction, tighter auth boundaries). If you run OpenClaw locally for automation, this is the kind of release you want: fewer silent failures, fewer sharp edges, and better visibility into usage.

Whats new in v2026.2.6

1) New model support: Opus 4.6 + GPT-5.3 Codex

The headline: OpenClaw now supports:

• •Anthropic Opus 4.6
• •OpenAI Codex gpt-5.3-codex (with forward-compat fallbacks)

That matters if youre running scheduled automation (cron jobs) or coding workflows and want the best available agentic model without manually hacking model IDs.

2) New provider: xAI Grok

OpenClaw added xAI (Grok) support as a provider. If your workflow spans multiple models (e.g., cheap model for triage, strong model for synthesis), this expands routing options.

3) Web UI: token usage dashboard

A token usage dashboard makes it easier to answer questions like:

• •Which jobs are costing me the most?
• •Which agent runs are inefficient?
• •Did last nights automation spike usage?

For teams running lots of background jobs, this is essential hygiene.

4) Memory: native Voyage AI support

If youre building workflows that need long-term recall (projects, customers, recurring context), native Voyage AI memory support is a notable addition.

5) Sessions: capped sessions_history payloads

One common failure mode in agent systems is context overflow. OpenClaw now caps sessions_history payloads to reduce overflow risk and stabilize agent turns.

Fixes you actually feel

Cron reliability fixes

If youre using OpenClaw to work while you sleep, cron reliability is the whole game. This release includes multiple cron fixes:

• •scheduling / reminder delivery regressions
• •next-run recompute hardening
• •timer re-arming
• •legacy schedule field support

In practice, this should reduce runs that say OK but didnt do the thing.

Security hardening: safety scanner + credential redaction

OpenClaw added a skill/plugin code safety scanner and also redacts credentials from config.get gateway responses. These are two high-leverage improvements because most real-world compromise paths are:

• •secrets leaking into logs / outputs
• •untrusted code getting executed with broad access

Canvas / A2UI auth requirements

Gateway canvas hosts and A2UI assets now require auth, reducing exposure if something is misconfigured.

What to do next (recommended)

If you run OpenClaw in production-like setups (VPS, always-on desktop, multiple integrations):

1) Update to v2026.2.6

2) Audit your skills/plugins and remove anything you dont use

3) Prefer isolated cron jobs for background work + keep delivery summaries concise

4) Verify cron jobs you care about by checking the run history regularly

Sources

• •OpenClaw v2026.2.6 release notes (GitHub): https://github.com/openclaw/openclaw/releases/tag/v2026.2.6

---

If you want, I can also publish a follow-up post showing a bulletproof OpenClaw cron setup (nightly research  write blog  build  push) with guardrails and failure alerts.