Claude Haiku 5.5 Free Tier and Claude Code: Limits and Setup

Update (9 Nov 2026): Claude Fable 5.2 and Mythos 5.2 are out - prices, access and benchmarks are in my Claude Fable 5.2 guide.

👁 115 reads

Claude Haiku 5.5 is available for free in the Claude app, but not in Claude Code - there you need a paid plan or API billing alongside version 2.1.293 or newer. Below is a step-by-step guide: how to enable it in both environments, how usage limits work, and when Haiku is enough instead of Sonnet 5.5. Pricing, benchmarks, and my test results in Polish are detailed in the Claude Haiku 5.5 guide.

Claude Haiku 5.5 for Free: How to Enable in the App

Anthropic explicitly states that Haiku 5.5 is available to users on Free, Pro, Max, Team, and Enterprise plans, via both web and mobile apps. You can set up a free account with an email address or a Google account. Then:

1. Open a new conversation on claude.ai or in the Claude app.
2. Click the model selector next to the prompt input box.
3. Select Haiku 5.5 from the list.
4. Type your prompt as usual - the selection applies to that specific conversation, so verify it when opening a new one.

Anthropic has announced no regional restrictions for Haiku 5.5. The API has no free tier - every request is billed starting from the first token ($0.10 / $0.50 per million tokens).

Free Tier Limits Explained

Anthropic does not disclose specific message counts. According to their help center:

• The Free plan limit operates on rolling sessions that reset every 5 hours.
• How much you fit into a single session depends on server load, conversation length, attachments, model choice, and reasoning effort.
• Web search and reading web pages consume allowance too - a long article supplied via link is ingested entirely into context.
• Before running out, you will see an “Approaching 5-hour limit” warning, and once exhausted, the exact reset time.

Does Haiku consume your limit slower than Sonnet? The help center notes usage varies by model and recommends Haiku as the most resource-efficient option, though without publishing an exact multiplier. Honestly: exact numbers are unavailable. The best strategies to make your free limit last longer:

• Start a new conversation for each distinct topic rather than stretching one thread across the entire day.
• Paste only the relevant excerpt of a document instead of the whole file.
• Use Haiku for straightforward questions, reserving Sonnet for more complex problems.

Anthropic states that Pro provides at least 5 times more usage per session compared to the Free tier.

How to Enable Haiku 5.5 in Claude Code

Claude Code is not included in the free tier. You need a Pro, Max, Team, or Enterprise subscription, or an API account in Claude Console. Steps to set it up:

1. Update Claude Code (claude update) and verify the version: claude --version. Haiku 5.5 requires 2.1.293 or higher.
2. Inside an active session, run /model and pick Haiku from the list, or type /model haiku directly. Pressing Enter saves it as default, while typing s switches it only for the current session.
3. From the terminal: claude --model claude-haiku-5-5.
4. Permanently: in ~/.claude/settings.json set "model": "haiku" or set the environment variable ANTHROPIC_MODEL=haiku.
5. Adjust thinking effort using /effort low (options: low, medium, high, xhigh, max; defaults to medium). You cannot disable reasoning entirely in Haiku 5.5.

Note on cloud platforms: the alias haiku targets Haiku 5.5 on Anthropic’s native API. On Amazon Bedrock, Google Cloud, and Microsoft Foundry, according to guidance from TestMu AI, it may still point to Haiku 4.5 - specify the full identifier there or set ANTHROPIC_DEFAULT_HAIKU_MODEL.

Best Use Case for Haiku: Subagents

Anthropic’s Claude Code documentation specifically recommends offloading lightweight subagent tasks to Haiku. Keep the main session on Sonnet or Opus, while tasks like searching files, parsing log outputs, and running test suites run on the cheaper model. Two ways to configure this:

• In the subagent definition file inside .claude/agents/, specify model: haiku in the header.
• Globally for all subagents: in settings.json under the env block, set "CLAUDE_CODE_SUBAGENT_MODEL": "haiku". To override individual subagent files, add "CLAUDE_CODE_SUBAGENT_MODEL_FORCE": "1".

The built-in Explore subagent inherits the main conversation model by default, so without these configuration tweaks it will not switch to Haiku automatically.

Claude Code Usage Limits

• Claude Code and the Claude web/mobile app pull from the same quota: a 5-hour rolling session limit plus a weekly limit.
• The message “You’ve hit your session limit” or “weekly limit” applies across all models - switching to Haiku will not bypass it; you must wait for the reset or purchase extra credits.
• A model-specific cap (“Opus limit”, “Sonnet limit”) is different - running /model haiku allows you to continue working immediately.
• Running /usage displays your quota progress bars alongside consumption by subagents, plugins, and MCP servers.
• The monthly API credit included with Max ($100 on Max 5x, $200 on Max 20x) does not apply to interactive Claude Code sessions - it works solely for API, Agent SDK, and Managed Agents calls.

When Is Haiku 5.5 Enough Instead of Sonnet 5.5?

Baseline numbers: Sonnet 5.5 costs $2 / $10 per million tokens on the API, which is 20 times more expensive. In terminal coding benchmarks (Terminal-Bench 4.0), Sonnet hits 70.6% versus Haiku’s 39.2%; in computer use (OSWorld 2.1), it reaches 83.9% versus 72.4%. In our Polish writing evaluation, Sonnet scored 80.4 versus 76.8 points (we tested it in Polish).

My rule of thumb: start with Sonnet or Opus for the primary interactive session, while assigning Haiku to subagents. If an entire session involves simple, repetitive code edits, switch the main session to Haiku and drop effort to low. If Haiku makes the same mistake twice in a row, switch back to Sonnet - fixing hallucinated loops costs more than the price gap.

Frequently Asked Questions

Is Claude Haiku 5.5 completely free?

Yes, inside the Claude web and mobile apps on the Free plan, subject to the standard 5-hour rolling limit. It is not free in Claude Code or via the API.

Can I run Claude Code on the free tier?

No. Claude Code requires a Pro, Max, Team, or Enterprise subscription, or a prepaid Claude Console API account.

How do I verify which Haiku version Claude Code is running?

Run /model to inspect the active model. On Anthropic’s direct API, the haiku alias resolves to Haiku 5.5; on third-party cloud platforms, provide the complete identifier claude-haiku-5-5.

Is Haiku 4.5 still operational?

Yes. As of October 8, 2026, it was marked active with a guaranteed operational window until “no earlier than October 15, 2026”, and Anthropic provides at least 60 days of advance notice before model retirement.

Google · Your sourcesQuantslant higher in your Google - one clickAdd to preferred sources →

Sources

• Anthropic, Claude Haiku - availability across plans and Claude Code.
• Anthropic, Claude Haiku 5.5 Documentation - model IDs, effort parameter, and API pricing.
• Claude Code, Model Configuration - aliases, /model command, minimum version requirements.
• Claude Code Subagents and Costs and Limits.
• Claude Plan Pricing, Usage and Length Limits Explained, Monthly API Credits for Max and Team Plans.
• TestMu AI on Haiku 5.5 in Claude Code - third-party cloud alias behavior.

• Read more about the CLI tool in our guide to Claude Code
• or see how the model compares to low-cost alternatives in Haiku 5.5 vs GPT-6 Luna vs Gemini 3.8 Flash. Verified October 8
• 2026