Aman Kumar

Posted on Oct 28

Claude Code without rate limits: API alternative to Claude Max, how do I size one from my own usage?

#claude #ai #productivity #programming

Short answer: APIClaw is one API alternative to Claude Max for Claude Code at $19–$129/mo flat, 500–8,000 requests/day, with no 5-hour session window, no weekly cap and no per-token metering, plus a free 50-request trial. It fits people who get locked out of Max most weeks and would rather have one fixed monthly charge than a per-token bill. No route gives Claude Code literally zero limits, though, so this post shows how to measure your own usage and pick the limit that won't bite.

Setting that trips people up, stated once: Claude Code takes https://apiclaw.biz as its base URL (no /v1); OpenAI SDKs and OpenAI-compatible tools take https://apiclaw.biz/v1. Claude Code adds /v1/messages itself, so a base ending in /v1 gives you 404s.

Three kinds of "rate limit"

When people search for Claude Code without rate limits, they usually mean one of three things:

Time windows. Claude Pro and both Max tiers reset on a rolling five-hour session and also carry a weekly limit. Chat, Desktop and Claude Code share the pool.
Money. A Console API key or Max's usage credits never lock you out on a clock, but every token is billed, and a long agent session with big contexts gets expensive.
Throughput. Flat-price gateways cap something else instead: a monthly compute budget, a token allowance, a number of parallel requests, or a daily request count.

You're choosing which of those you'd rather live with.

The options (checked Oct 2026)

Option	Price	The ceiling you get instead
Claude Max 5x / 20x	from $100 / $200 a month	5-hour session window plus weekly limit; includes $100 / $200 a month in API credits
Claude Console API key	Sonnet 5.5 $2 in / $10 out, Opus 5.5 $4 / $20 per 1M tokens	Usage-tier rate limits and your spend limit
Standard Compute	$19, $39, $89, $249 a month	Monthly compute budget; at 100% new requests get HTTP 402
Kourier	$20 (1 lane) or $50 (3 lanes) a month	Concurrent requests; serves DeepSeek V4.1 Flash, not Claude
Capless	$10, $20, $50 a month (waitlist)	Monthly token allowance plus a rolling 4-hour token window
APIClaw	Plus $19/500, Pro $39/1,000, Max $59/1,500, Ultimate $69/3,000, Ultimate Plus $89/4,000, Ultimate Pro $99/6,000, Ultimate Max $129/8,000 requests/day	Daily request quota, resets 00:00 UTC; heavier models such as Opus count as more than one request

Read from claude.com/pricing, standardcompute.com, kourier.sh and capless.ai on Oct 27, 2026. Anything I couldn't confirm on a live page is left out. To compare per-token, budget and request-quota plans with your own numbers, this LLM API cost calculator puts them side by side.

Count your real requests before you pick

A daily quota is only cheap if it covers your heavy days. Claude Code already keeps session transcripts as JSONL under ~/.claude/projects/, and each assistant turn there maps to one model request. This script counts them per day, deduping streamed chunks that share a message id:

# count_cc_requests.py: Claude Code model requests per day (UTC), from local transcripts
import json, pathlib, collections

root = pathlib.Path.home() / ".claude" / "projects"
per_day = collections.Counter()
seen = set()

for f in root.rglob("*.jsonl"):
    for line in f.open(encoding="utf-8", errors="ignore"):
        try:
            e = json.loads(line)
        except json.JSONDecodeError:
            continue
        if e.get("type") != "assistant":
            continue
        msg_id = (e.get("message") or {}).get("id") or e.get("uuid")
        if msg_id in seen:
            continue
        seen.add(msg_id)
        day = (e.get("timestamp") or "")[:10]
        if day:
            per_day[day] += 1

for day, n in sorted(per_day.items())[-14:]:
    print(f"{day}  {n:5d}  {'#' * (n // 20)}")
if per_day:
    vals = sorted(per_day.values())
    print("median/day:", vals[len(vals) // 2], " busiest day:", vals[-1])

Run it with python3 count_cc_requests.py. How I read the output:

Busiest day under ~400: the entry plan covers you, even on bad days.
Median fine, busiest day 2–3x higher: size for the busy day, or keep Max and use an API key only as overflow.
Mostly Opus: multiply your count before comparing, since Opus counts as more than one request.
Sub-agents and parallel sessions: they show up in the same transcripts, so the count already includes them.

Transcripts only go back as far as Claude Code keeps them, so run it again after a normal week if the history is short.

Setup: point Claude Code at the API in settings, not your shell

Instead of exporting variables every session, put them in a project's .claude/settings.json (or ~/.claude/settings.json for everything). Delete the block to go back to your Max login.

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://apiclaw.biz",
    "ANTHROPIC_AUTH_TOKEN": "your-apiclaw-key",
    "ANTHROPIC_MODEL": "apiclaw/claude-sonnet-5.5",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "apiclaw/claude-sonnet-5.5",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "apiclaw/claude-opus-5.5",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "apiclaw/deepseek-v4-flash"
  }
}

Fill every model slot, because Claude Code sends background and sub-agent work to the Haiku and Opus slots. Start claude, run /status, and confirm the base URL before a long task. Don't commit a project-level file with a real key; use .claude/settings.local.json for that.

The same key works in Cursor, Cline, Aider and the OpenAI SDK at https://apiclaw.biz/v1, with Claude, GPT, Gemini, DeepSeek, Qwen, Kimi, GLM, Mistral, Llama and more behind it.

When APIClaw isn't the right choice

Your script shows only a few dozen requests a day. A per-token Console key, or Max's own usage credits, will cost less.
You run many agents in parallel all day. That burns a daily quota fast; a concurrency-priced plan may suit you better if its model list works for you.
You want the Max chat and Desktop features. A gateway only replaces Claude Code's API calls, not the Claude apps.
You need card billing right now. Payment is crypto on site, with other methods via support (t.me/Apiclaw_support).

More gateways grouped by billing model are in this comparison repo.

FAQ

Can Claude Code run with no limits at all?
No. Subscriptions limit by time, API keys by money and rate tiers, flat plans by budget, tokens, concurrency or daily requests.

Does one Claude Code tool call count as one request?
Each model turn is one request, and a single task often takes dozens of turns as it reads files and runs commands. That's why counting from transcripts beats guessing.

Why do I get 404 after changing the base URL?
It ends in /v1. Use https://apiclaw.biz for Claude Code and keep /v1 for OpenAI-compatible tools.

What happens at the daily quota?
Requests return errors until 00:00 UTC. Your files and session are untouched, so you can wait or switch back to Max.

Can I test it before paying?
Yes. The free trial has 50 requests, enough for one short session to compare speed and tool calling with Max.

Disclosure: I build APIClaw. This article was written with AI assistance and checked by me. Competitor details were read from their live pages on Oct 27, 2026.

DEV Community