DeepSeekV4 + Claude Code = 100X Cheaper
The video demonstrates how to combine Claude Code with DeepSeek V4 to reduce AI coding costs by up to 100x. The presenter walks through setting up a proxy repo that lets developers run Claude Code alongside DeepSeek simultaneously, using each model for its strengths: Claude for UI/design and DeepSeek for backend/algorithmic work.
Summary
The video opens with host Jack Roberts, a founder who built and sold a startup to 60,000 customers, introducing a method to use the Claude Code framework with DeepSeek V4 at a fraction of the cost. He highlights that DeepSeek V4 has 1.6 trillion parameters, a 1 million token context window, and is MIT open-weights, making it compatible with the Claude Code ecosystem. A key selling point is its strong tool-calling ability, which mirrors the MCP workflow used with Anthropic models.
On pricing, Roberts compares Claude Opus at roughly $62.50 per session versus DeepSeek V4 at dramatically lower rates, with a promotional period making it even cheaper. DeepSeek V4 Flash is noted as nearly free at around $0.20 for a four-hour session. He references benchmark comparisons (SWE-Bench, Terminal Bench, MCP Atlas) showing DeepSeek V4 competing closely with Claude Sonnet 4.6 and approaching Opus 4.7 performance levels.
The setup tutorial centers on a GitHub repo that acts as a proxy server intercepting Claude API calls, enabling a 'dual terminal' workflow where DeepSeek, free models, and Claude Code can run in separate terminal windows simultaneously within an IDE like Cursor (referred to as 'anti-gravity'). Roberts walks through cloning the repo, adding a DeepSeek API key from platform.deepseek.com (recommending a $2–$5 deposit over Open Router due to severe rate limiting on the free tier), and creating launch scripts with and without the dangerous-skip-permissions flag.
To demonstrate practical capability, Roberts tests DeepSeek's tool/skill-calling by shortening a URL via a Bit.ly MCP skill — confirming it works reliably. He then builds a coffee-themed website using a division-of-labor approach: Claude handles the creative design using the 'Awesome Designer MD' design system repo and an Apple design framework, while DeepSeek is used to build a sophisticated multi-variable productivity ROI calculator added to the site. Roberts notes that DeepSeek successfully replicated the existing HTML style when given context, but struggled with original creative design work.
He also covers using Open Router with free cycling models as a zero-cost alternative, and describes extending the pipeline further by tagging in GPT-4.5 Codex (via an existing ChatGPT Plus subscription) for code review, creating a three-model workflow: Claude for design, DeepSeek for heavy lifting, and Codex for review.
The video closes with clear guidance on when NOT to use DeepSeek: corporate environments, healthcare, finance, or any highly regulated/sensitive IP contexts, due to data privacy concerns with the model's Chinese origin. For solo devs, open source projects, prototyping, and learning Claude Code workflows, Roberts strongly endorses it as a cost-reduction tool.
Key Insights
- Roberts argues that DeepSeek V4's most compelling feature for this use case is not raw benchmark performance but its tool-calling ability, which mirrors the MCP workflow used with Anthropic — making it a functional drop-in for non-visual coding tasks rather than just a cheaper alternative.
- Roberts found that using DeepSeek via Open Router is 'basically unusable' due to severe rate limiting, and recommends depositing $2–$5 directly on platform.deepseek.com to get unrestricted API access instead.
- Roberts claims DeepSeek V4 lacks 'creative imagination' for original UI/UX work but performs well when adding features within an already-established design framework — meaning it replicates style effectively but cannot originate it.
- Roberts explicitly warns against pasting API keys, naked URLs, or sensitive data into DeepSeek, stating that anything entered into the chat should be treated as potentially public — and advises corporate, healthcare, and finance users to avoid the model entirely.
- Roberts describes a three-model production pipeline — Claude for design, DeepSeek for algorithmic/backend heavy lifting, and GPT-4.5 Codex (leveraged via an existing ChatGPT Plus subscription at no extra cost) for code review — as the optimal cost-to-quality workflow.
Topics
Transcript
[0:00] Imagine if you could use Claude Code 100 times cheaper. In this video, I'm going to show you exactly how to combine the Claude Code framework with the brand new DeepSeek V4. So, you can save thousands of dollars, build as much as you want with no rate limitations. And with this brand new repo, set it up in minutes, even if you're starting as a complete beginner. And if you don't know who I am, my name is Jack Roberts. I built and sold my last tech startup over 60,000 customers. Now, I'm building my own AI companies. and I share the stuff here that actually works. So, if you haven't [0:31] already, grab that beautiful coffee and…
Full transcript available for MurmurCast members
Sign Up to AccessMore from Jack Roberts
100 hours of Hermes Agent lessons in 23 minutes
Jack walks through advanced features of Hermes Agent, an AI personal assistant, covering memory systems, background tasks, scheduled cron jobs, model switching, and integration with tools like Obsidian, GitHub, and various AI models. The video aims to help users unlock capabilities beyond basic chatbot usage. Key themes include connecting external memory systems, delegating tasks to specialized AI models, and building a persistent, context-aware personal assistant.
Claude Code = $10,000 Beautiful Slides
Jack demonstrates a GitHub-based system using Claude Code to generate professional slide decks from any website URL by codifying 20 universal design principles. The system uses Firecrawl to extract brand DNA from websites and can integrate AI-generated images via APIs like Krea.ai. The result is polished, brand-consistent HTML presentations created in minutes with minimal input.
The most valuable thing I got from working at McDonald's
The speaker reflects on their first job at McDonald's at age 17, earning £5 an hour. The most valuable takeaway was the realization that they never wanted to work such a job again, reinforced by calculating how long it would take to earn a million pounds at that wage.
Claude Code Memory System = CHEAT CODE
Jack Roberts presents a three-tier Claude memory system designed to give AI tools persistent context across all platforms and sessions. The system consists of short-term identity memory, mid-term project memory via structured folders and claude.md files, and long-term memory using either Obsidian or Pinecone for archiving conversations and expert knowledge. The goal is to eliminate the 'amnesia' problem where AI loses context between chats.
ChatGPT ads aren't actually a bad thing. Here's when I want them.
The speaker argues that ads in ChatGPT are not inherently bad, particularly when users are actively seeking product recommendations. They distinguish this as 'pull advertising,' where the user initiates the request and ChatGPT synthesizes results based on provided context.