Model help

[lethal tusk]

Which model is the best kimi k2.5,Gemini 3 pro or minimax m2.5

[zealous lily]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[subtle prawn]

There isn’t a single “best” — it depends what you’re optimizing for (quality, tool-calling reliability, speed, cost, multimodal).

Here’s a practical way to choose between Kimi K2.5 vs Gemini 3 Pro vs MiniMax M2.5:

• If you need vision / multimodal (images, screenshots) or very strong general reasoning: Gemini 3 Pro is usually the safest bet.
• If you want strong value (good quality per $) and fast general chat/brainstorming: MiniMax M2.5 is often a great “daily driver” choice.
• If your workload is heavy on long-context reading/summarizing and you like the Kimi style: Kimi K2.5 is often picked for that niche.

For OpenClaw specifically: “best” often means most reliable at tool calling (browser, exec, structured outputs). If you’re using lots of tools, pick the one that behaves most consistently in your setup and prompts—model behavior can vary a lot by provider settings.

Quick test I’d run:

1. a tool-heavy task (browse → extract → summarize)
2. a structured-output task (JSON schema / config snippet)
3. a long-context task (paste 30–100k text and ask targeted questions)
   Compare: failure rate + cost + latency.

OpenClaw model notes: https://docs.openclaw.ai/concepts/models

[lethal tusk]

I'm on a vps how do I get my agent to use browser and all also latency is quite bad with all 3 of them..

[subtle prawn]

On a VPS, “agent can use browser + web tools” basically comes down to (1) enabling the tools in config and (2) having a browser it can actually control (usually headless).

1) Browser tool on a VPS (recommended: OpenClaw-managed headless)

Make sure browser is enabled + switch default profile to the managed one:

{
  browser: {
    enabled: true,
    defaultProfile: "openclaw",
    headless: true
  }
}

Then test from the VPS shell:

openclaw browser status
openclaw browser start
openclaw browser open https://example.com
openclaw browser snapshot

Docs: https://docs.openclaw.ai/tools/browser
Headless note: https://docs.openclaw.ai/help/faq#can-the-openclaw-browser-run-headless

Common gotchas on VPS:

• Playwright not available → browser actions/snapshots will fail until Playwright/Chromium deps are installed (the error message usually says this explicitly). See the troubleshooting section in the browser doc.
• Some hosts need browser.noSandbox: true (only if you see sandbox-related Chromium errors).

2) Web search / web fetch

web_fetch works without keys. web_search needs a Brave Search key:

{
  tools: {
    web: {
      search: { enabled: true, apiKey: "BRAVE_API_KEY_HERE", maxResults: 5 },
      fetch: { enabled: true }
    }
  }
}

Docs: https://docs.openclaw.ai/help/faq#how-do-i-enable-web-search-and-web-fetch

3) If you actually want to browse “on your laptop” while Gateway stays on the VPS

Pair your laptop as a node, and let the Gateway proxy browser actions to that machine (best if you want a real, non-headless browser / extension takeover). Docs: https://docs.openclaw.ai/nodes

---

Why latency feels bad (even if the models are “good”)

A few super common causes:

1. Region mismatch: VPS in EU but provider endpoint/model routes to US/Asia (or vice versa). Try a provider/endpoint that’s geographically close to your VPS.
2. Fallbacks/retries: if auth is flaky or rate limits hit, it may silently try multiple providers/models before succeeding. Run:
   • openclaw models status
   • openclaw logs --follow
3. Tool round-trips: browsing + snapshot + act is multiple network calls; keep browser snapshots “efficient” and avoid extra steps when possible.

If you paste (a) openclaw models status and (b) the output of openclaw browser status (redact tokens), I can usually pinpoint whether it’s “browser not installed right” vs “provider routing/latency”.

[lethal tusk]

What's the best tool I can give it to make me pdf's

[subtle prawn]

If your goal is “turn something into a PDF and send it back in chat”, the best built-in tool is the Browser tool — it has native PDF export (Playwright “print to PDF”).

Best option: browser → pdf

Use it for:

• saving a webpage as PDF (receipts, docs, articles)
• generating a nice-looking PDF by rendering HTML and printing it

Docs: https://docs.openclaw.ai/tools/browser

If you mean “generate PDFs from text/markdown”

Then give it exec (or a paired node with system.run) and install a converter like:

• pandoc (markdown → pdf)
• typst (great templates, fast)

Workflow: agent writes report.md → runs pandoc report.md -o report.pdf → sends it.

One important gotcha (why PDFs sometimes “don’t send”)