Anthropic · checked October 13, 2026

How do I run Claude Haiku 5.5 in Claude Code, OpenCode or Codex?

Q: How do I run Claude Haiku 5.5 in Claude Code, OpenCode or Codex?

Claude Haiku 5.5 runs on Standard Compute at https://api.stdcmpt.com/v1 with the model ID anthropic/claude-haiku-5.5. Check its price step at 100,000 prompt tokens before you make it your main coding model.

Claude Haiku 5.5 runs on Standard Compute at https://api.stdcmpt.com/v1 with the model ID anthropic/claude-haiku-5.5. Check its price step at 100,000 prompt tokens before you make it your main coding model.

Model	Anthropic's fastest Claude 5.5 model, aimed by Anthropic at high-volume, latency-sensitive work such as classification, extraction and routing. It has a 1M-token context window and 128K max output.
Model ID	`anthropic/claude-haiku-5.5` Short form `claude-haiku-5.5` also works
Base URL	`https://api.stdcmpt.com/v1` Claude Code uses `https://api.stdcmpt.com`
Context	1M tokens; about 850K input and 50K output per request
Input	Text, images and files
Reasoning	low, medium, high, xhigh or max, set per model on the Models page

Copy a working config

Haiku 5.5 works in all three agents. It earns its place on short prompts, so it suits subagents, scripted checks and an OpenCode or Codex session on a small repo.

terminal/v1/messages

# Install the Standard Compute launcher once
curl -fsSL https://standardcompute.com/cli/install.sh | sh

# Start Claude Code on your key, pinned to anthropic/claude-haiku-5.5
standardcompute launch claude -- --model anthropic/claude-haiku-5.5

# Or configure it yourself. No /v1 here: Claude Code adds /v1/messages.
export ANTHROPIC_BASE_URL=https://api.stdcmpt.com
export ANTHROPIC_AUTH_TOKEN=YOUR_API_KEY
export ANTHROPIC_MODEL=anthropic/claude-haiku-5.5
claude

{
  "$schema": "https://opencode.ai/config.json",
  "model": "standardcompute/anthropic/claude-haiku-5.5",
  "provider": {
    "standardcompute": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Standard Compute",
      "options": {
        "baseURL": "https://api.stdcmpt.com/v1",
        "apiKey": "{env:STANDARDCOMPUTE_API_KEY}"
      },
      "models": {
        "anthropic/claude-haiku-5.5": { "name": "Claude Haiku 5.5" },
        "standardcompute": { "name": "Smart routing" }
      }
    }
  }
}

# Install Codex and the Standard Compute launcher once
npm install -g @openai/codex
curl -fsSL https://standardcompute.com/cli/install.sh | sh

# Start Codex pinned to anthropic/claude-haiku-5.5
standardcompute launch codex -- --model anthropic/claude-haiku-5.5

Replace YOUR_API_KEY with a key from the dashboard; OpenCode reads it from STANDARDCOMPUTE_API_KEY. Full setup: Claude Code, OpenCode, Codex CLI.

What it costs

Per 1M tokens	Prompt up to 100K	Prompt over 100K
Input	$0.10	$0.50
Cache hit	$0.01	$0.05
5-minute cache write	$0.125	$0.625
Output	$0.50	$2.50

Anthropic standard API rates. Prompt length counts cache reads too, and a request over the line pays the higher rate on all of its tokens. Source: Anthropic API pricing, checked October 13, 2026.

A 150K-token turn on Haiku 5.5, past the 100K line

19,500 fresh input × $0.50/M$0.0097

130,500 cache hits × $0.05/M$0.0065

700 output × $2.50/M$0.0018

One turn$0.0180

Averages from coding-agent traffic on our gateway, Sep 13 to Oct 12, 2026: 87% of prompt tokens are cache hits and a turn returns about 700 tokens. Vendor list prices. The same turn at 90K tokens costs $0.0023.

Under 100K tokens Haiku 5.5 is the most efficient Claude model you can pin, and the $20 budget in the $19 Starter plan covers thousands of short turns. Past the line, the same turn costs about eight times as much.

What works differently here

Coding agents cross the 100K line constantly. In a month of coding-agent traffic on our gateway, about half of all requests had prompts over 100,000 tokens, and each of those pays five times the short-prompt rate. Priced on that traffic, Haiku 5.5 cost 3.1 times as much as GPT-6 Luna.
Claude Code's haiku shortcut is smart routing on Standard Compute, not Haiku 5.5. Use the full ID anthropic/claude-haiku-5.5 to pin it.
The Messages endpoint ignores Claude Code's thinking setting and returns no thinking blocks; set the reasoning level under Customize on the Models page.

Other ways to run Claude Haiku 5.5

Anthropic API From $0.10 / $0.50 per 1M	First-party, pay per token. Batch requests are 50% off and prompt caching is yours to set up.
Claude Pro / Max Subscription	Claude Code's native login. Usage runs in session and weekly limits rather than a token bill.
Bedrock / Vertex / Foundry Pay per token	Cloud billing for procurement teams. On Bedrock the ID is anthropic.claude-haiku-5-5.
Standard Compute Monthly plan	One key for Claude Code, Codex and OpenCode, drawing on a monthly compute budget.

FAQ

Is Claude Haiku 5.5 a good main model for Claude Code?

Only if your sessions stay small. Claude Code's prompt grows with every step, and once it passes 100,000 tokens Haiku 5.5 bills at $0.50 / $2.50 per million. For long sessions, GPT-6 Luna or a routed selection is more efficient.

Claude Haiku 5.5 or GPT-6 Luna?

Both list at $0.10 / $0.50 per million tokens on short prompts. Luna keeps that rate until 272K tokens, Haiku 5.5 only until 100K. We priced a month of real coding-agent traffic on both; the comparison is linked below.

Plans from $19/mo

One key for Claude Haiku 5.5 and the rest of the catalog, with a monthly compute budget you can watch in the dashboard and no overage charges.

Get your API key

Sources: Anthropic API pricing · Anthropic models overview
Related: Claude Haiku 5.5 vs GPT-6 Luna, priced on real traffic · Run GPT-6 Luna · All model IDs