AI Models

DeepSeek

Open-source frontier reasoning, via BYOK on OpenRouter.

DeepSeek V4 and V4 Pro are open-source models that punch in the same weight class as Claude Sonnet and GPT-5 on reasoning benchmarks, at a fraction of the price. MoClaw doesn't include DeepSeek in your monthly credits, but you can route every call through your OpenRouter key and pay provider rates directly with no markup.

無料で試す すべての integrations

仕組み

engineering なしで 3 steps で DeepSeek を接続します。

  1. 1

    Top up an OpenRouter balance

    Go to openrouter.ai, add $10 to your account. OpenRouter passes through DeepSeek's pricing — currently around 90% cheaper than Claude Sonnet for the same task.

  2. 2

    Paste your OpenRouter key into MoClaw

    Settings then Models then BYOK then OpenRouter. Stored encrypted at rest, decrypted only at request time.

  3. 3

    Pick DeepSeek V4 or V4 Pro in the model picker

    Switch on a per-chat basis. The rest of MoClaw — skills, browser control, cloud computer, scheduled jobs — keeps working the same.

こう言ってみる

DeepSeek に貼り付けられる実際の prompts です。

  • › On DeepSeek V4 Pro, walk through this 2,000-line algorithm and find the off-by-one bug.
  • › On DeepSeek V4, summarize these 50 customer interview transcripts into 5 themes.
  • › On DeepSeek V4, classify 10,000 support tickets in batch — I care about cost more than quality.

Step by step demo

prompt を送った後に実際に起きること。

Prompt 01 4 steps

“On DeepSeek V4 Pro, walk through this 2,000-line algorithm and find the off-by-one bug.”

MoClaw がすること

  1. 1 Reads the file top-to-bottom in one pass (V4 Pro has a 128K context window).
  2. 2 Builds a data-flow trace through the loop in question.
  3. 3 Flags line 1,847 as the off-by-one: index runs i ≤ n.len, should be i < n.len.
  4. 4 Suggests a 1-line fix and the failing test case that would catch it.
結果

Returns a code review with the bug pinpointed at line 1,847, a 1-line patch, and a Go test that fails before and passes after. Costs about $0.04 on OpenRouter — roughly 4 credits if it had been Claude Sonnet.

これらの workflows 向け

ユーザーが MoClaw 上で DeepSeek によって自動化する workflows。

Use case

AI Agents Research Digest

MoClaw monitors arXiv daily for the latest AI agent papers, summarizes the top findings, and delivers a curated research digest on your schedule.

Use case

Automated Email Auto-Reply for Business Inquiries

MoClaw monitors your inbox for cooperation and quotation requests, auto-replies with your pricing, and notifies you when someone accepts.

FAQ

pricing、privacy、limits についての短い回答。

Why isn't DeepSeek in the included plan?
OpenRouter is a third-party gateway and DeepSeek is open-source. Bundling it would mean MoClaw eats the variable cost. BYOK keeps you paying provider rates directly with no MoClaw markup.
How does DeepSeek compare to Claude Sonnet on real tasks?
On reasoning benchmarks (MMLU, MATH, HumanEval) V4 Pro is within ~5% of Sonnet 4.6 at roughly 1/10 the cost. On creative writing and complex multi-step agentic work, Claude is still ahead. Try both for your own workload.
Can I use DeepSeek with skills, browser control, cron?
Yes. BYOK only changes the model billing. The MoClaw runtime — skills, browser control, cloud computer, scheduled jobs, file handling — works identically.
What about V4 vs V4 Pro?
V4 is fastest and cheapest, comparable to Haiku in speed. V4 Pro adds reasoning ability, comparable to Sonnet. Both are available on OpenRouter.

MoClaw を無料で試す。

毎月 1,000 credits、または 自分の key を持ち込むことで無制限に利用できます。

無料トライアル開始

いつでも解約