I'll show you how to cut your coding assistant costs dramatically — without giving up strong models or the Claude Code workflow you're already used to.

Claude Code has quickly become the developer's tool of choice when it comes to AI-powered coding assistants. Most of us have always gone for one of Anthropic's subscriptions to get access to the models Sonnet or Opus. But this comes with a heavy price tag — up to $200 per month, which is hard to justify if you're not running agents 24/7.

Why pay $200 a month when your LLM doesn't even remember what you did yesterday?

So how can we cut down on those costs without losing quality in terms of the capabilities of the models?

1. Get the right subscription for your usage

Not every developer needs the Max plan. Anthropic's Pro subscription is often sufficient for daily coding tasks.

My recommendation: start with Pro. If you consistently hit rate limits and can't wait for them to reset, then — and only then — consider Max.

2. Use GLM 4.6 instead of Claude

I have already talked about it in this article, but I really want you to give this a try. GLM 4.6 is a surprisingly capable model when given sufficient context and clear instructions.

Considering how inexpensive the subscription is, the quality is much closer to Anthropic's models than you'd expect, and it's compatible with Claude Code as well. The linked article explains how to set it up for usage with Claude Code.

3. Use DeepSeek 3.2 instead of Claude

With the arrival of DeepSeek 3.2, the team behind DeepSeek released a model that is both highly capable and ridiculously cheap.

On SWE Benchmark, DeepSeek 3.2 performs only slightly behind Claude 4 Sonnet — at roughly one tenth of the cost.

Setup: Using DeepSeek 3.2 with Claude Code

  1. Create an account on the DeepSeek-Platform.
  2. Create an API-Key (e.g. CC-Lenovo-Windows) and copy it
  3. Add funds to your DeepSeek account (start out with $10)
  4. Install Claude Code (if you haven't already) curl -fsSL https://claude.ai/install.sh | bash
  5. Add your API-key to ~/.zshenv or~/.bash_profile : export DEEPSEEK_API_KEY="sk-xxxxxxxx"
  6. Add the following alias to your ~/.bashrc, ~/.zshrc, or ~/.profile:
alias claudeds='
ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic \
ANTHROPIC_AUTH_TOKEN="$DEEPSEEK_API_KEY" \
ANTHROPIC_MODEL=deepseek-chat \
ANTHROPIC_DEFAULT_HAIKU_MODEL=deepseek-chat \
CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1 \
API_TIMEOUT_MS=600000 \
claude
'

Pro tip: You can use ANTHROPIC_MODEL=deepseek-reasoner too for the thinking model for better planning performance, but beware that this model is slower.

7. Restart your terminal

8. Run claudeds in your terminal. Claude Code will now use DeepSeek's chat model (currently DeepSeek 3.2).

None
Keep piling on the money.

Conclusion

Claude Code doesn't have to be a $200-per-month commitment. With a bit of setup, you can keep the exact same workflow while dramatically reducing costs.

Whether you downgrade your Anthropic subscription, switch to alternatives like GLM 4.6, or fully replace Claude models with DeepSeek 3.2, the result is the same: more tokens, fewer invoices, and no noticeable drop in productivity.

In 2025, the smartest optimization isn't writing less code — it's paying less for the AI that writes it with you.

If you found this useful, follow me — I'm releasing content for different things regarding AI, software development, productivity, etc. Also feel free to comment and reach out to me.

Until next time!