sllm vs ANTHROPIC_BASE_URL: two ways to get unlimited Claude tokens

Two approaches just landed on Hacker News in the same week: sllm (split a GPU node with other developers) and ANTHROPIC_BASE_URL proxies. Both promise unlimited tokens. They work very differently.

Here's the practical comparison.

The core difference

sllm gives you shared compute — you're literally splitting GPU time with strangers on a node. Cheap, but your inference speed depends on who else is using the node right now.

ANTHROPIC_BASE_URL proxies give you a managed Claude API endpoint. You set one environment variable, and your Claude Code sessions route through it. You still get Claude's actual models, not a local alternative.

# The entire setup for ANTHROPIC_BASE_URL approach:
export ANTHROPIC_BASE_URL=https://simplylouie.com/api
export ANTHROPIC_API_KEY=your-key
claude  # now using proxy, rate limits lifted

That's it. One variable.

Side-by-side comparison

Feature	sllm	ANTHROPIC_BASE_URL proxy
Model	Local/open weights	Real Claude (Sonnet/Haiku)
Setup	Run local server	One env var
Rate limits	None (your GPU slice)	Lifted via proxy
Anthropic ToS	N/A	Compliant
CVE risk	Depends on impl	None (hosted)
Price	Varies	✌️$2/month
Works with Claude Code	Needs config	Native
Context window	Model-dependent	Claude native

When to use sllm

sllm makes sense when:

• You specifically want open weights (Llama, Mistral, etc.)
• You're doing batch inference and don't need Claude specifically
• You want to experiment with local models at scale
• You're comfortable managing shared infrastructure

When to use ANTHROPIC_BASE_URL

ANTHROPIC_BASE_URL proxy makes sense when:

• You're using Claude Code and hitting rate limits mid-session
• You need Claude specifically (not a local alternative)
• You want zero setup — one env var, done
• You're in a team where everyone needs the same setup

The Claude Code use case

This is where ANTHROPIC_BASE_URL wins completely. Claude Code expects Anthropic's API. The ANTHROPIC_BASE_URL environment variable exists specifically to redirect those calls:

# In your shell profile:
export ANTHROPIC_BASE_URL=https://simplylouie.com/api
export ANTHROPIC_API_KEY=sl-your-key-here

# Now run Claude Code normally:
claude "refactor this auth module"

# No rate limits. Same Claude. Session continues uninterrupted.

With sllm, you'd need to reconfigure Claude Code to point to a different model endpoint — and you'd lose Claude's specific capabilities.

The rate limit problem, specifically

Claude Code's rate limit hits mid-session. You're in the middle of a refactor. Claude stops:

Claude is currently unavailable. Please wait before continuing.

The ANTHROPIC_BASE_URL approach means that message never appears. The proxy manages the rate limiting transparently — you keep coding.

Real-world setup (30 seconds)

# 1. Sign up at simplylouie.com
# 2. Get your API key from dashboard
# 3. Add to .bashrc or .zshrc:
export ANTHROPIC_BASE_URL=https://simplylouie.com/api
export ANTHROPIC_API_KEY=sl-your-key

# 4. Source it:
source ~/.bashrc

# 5. Run Claude Code:
claude
# Done. Rate limits gone.

For CI/CD:

# .github/workflows/ai-review.yml
env:
  ANTHROPIC_BASE_URL: https://simplylouie.com/api
  ANTHROPIC_API_KEY: ${{ secrets.LOUIE_API_KEY }}

The cost comparison

• sllm: Variable (depends on GPU time purchased)
• Claude Pro: $20/month (with rate limits)
• SimplyLouie proxy: ✌️$2/month (rate limits lifted, real Claude)

For developers who specifically use Claude Code and just want the rate limits gone, the math is straightforward.

Bottom line

If you want open weight models at scale, sllm is interesting infrastructure.

If you want Claude Code to work without interruption, ANTHROPIC_BASE_URL is the correct tool. One env var. $2/month. Rate limits gone.

→ SimplyLouie ANTHROPIC_BASE_URL setup — 7-day free trial, no card required

---

Both approaches have their place. The question is whether you need Claude specifically or just need cheap inference tokens.