Claude Haiku 5.5 Explained: Free on Every Plan, and When to Actually Pick It

Anthropic has released another Claude model, and this one is the cheap fast one. On October 7, 2026 it launched Claude Haiku 5.5, completing the Claude 5.5 family after Opus 5.5 and Sonnet 5.5. Almost everything in the announcement is aimed at developers: token prices, benchmarks, subagents. The one line that matters to everyone else is on Anthropic's own Haiku page, and it is unusually blunt about who gets it: Free, Pro, Max, Team, and Enterprise users can select Haiku 5.5 on Claude.ai.

Key Takeaways

• It shipped October 7, 2026. Anthropic's help centre release notes carry the launch entry under that date, and the announcement page prints the same date in its own dateline.
• Free Claude gets it, and Anthropic says so in those words. Its Haiku page names "Free, Pro, Max, Team, and Enterprise" outright. Its Opus and Fable pages list the same plans with Free left out, so the wording is a real distinction.
• You have to pick it. Anthropic has not called Haiku 5.5 a default on any plan. Click the model name next to the send button, and then "More models" if it is not in the short list.
• It is the fastest Claude so far, with one caveat Anthropic itself writes. Its footnote: Haiku 5.5 is "our fastest model to date at each model's standard speed, although it runs less quickly than our Opus models in Fast Mode."
• It knows as much as the expensive models. Training data runs to June 2026, the same cutoff as Sonnet 5.5, Opus 5.5 and Fable 5.1, and eleven months past Haiku 4.5's July 2025.
• It is the first Haiku with an effort setting, so you can trade quality against how fast you burn through your usage limit. Thinking, though, cannot be switched off in the Claude apps.
• It watermarks its text from launch. Anthropic's marking table ticks Haiku 5.5 in all three columns.
• Anthropic does not claim it is the best model for your chat. Its own words: Sonnet 5.5 and Opus 5.5 "remain better choices for complex agentic coding tasks," and Haiku 5.5 "is best suited to more narrowly scoped tasks."
• If you pay for Max or Team, there is money attached to this release that you have to go and claim. Max 5x gets $100 a month in API credit, Max 20x $200, Team up to $500 pooled. Free and Pro get nothing.

Do You Get It? Availability by Plan

Every Claude plan, including the free one. Here is the picture as of October 12, 2026, read off Anthropic's Haiku model page and the claude.com/pricing plan matrix.

Why we can say Free gets it. Anthropic's four model pages share an "Availability and pricing" section, and the first sentence of each one draws the line itself. The Opus page says Opus 5.5 "is available on Claude for Pro, Max, Team, and Enterprise users"; the Fable page says Fable 5.1 "is available to Pro, Max, Team, and Enterprise users." Both list four plans and both omit Free. The Haiku page lists the same four plans with Free in front of them. That is a positive statement rather than marketing filler, and the pricing matrix agrees: the Haiku row reads Yes for Free, Pro, Max 5x and Max 20x, while the Opus row reads No for Free.

One honest wrinkle: the Haiku page's headline still reads "Claude Haiku 4.5" while the Availability section underneath is all about 5.5. Anthropic is slow to update those hero banners. The Availability section is the part that was rewritten for this launch, and it is the part the release notes point at.

What It Actually Is, and When to Pick It

A speed-and-cost model, not an upgrade to your everyday chat. Anthropic is clearer about this than vendors usually are. It says Haiku 5.5 "is designed for high-volume, cost-sensitive tasks" and "reliably handles quick and repetitive workloads (like summaries, compactions, database queries, and classification requests)," and then, in the same post, that "Sonnet 5.5 and Opus 5.5 remain better choices for complex agentic coding tasks" while Haiku 5.5 "is best suited to more narrowly scoped tasks that might otherwise have been cost-prohibitive with previous versions of Claude."

Translated into what you would actually do with it:

The benchmark gap behind that advice is wide in both directions. Haiku 5.5 scores 39.2% on Terminal-Bench 4.0 against Haiku 4.5's 0.0%, which is a huge generational jump, and against Sonnet 5.5's 70.6%, which is the gap that should decide your model menu. On Humanity's Last Exam without tools it scores 45.9% against Sonnet 5.5's 56.9%.

What It Knows, and How Much It Can Hold

June 2026, and a million tokens when you chat with it. Both numbers come from Anthropic's help centre, re-read October 12, 2026, with revision stamps from the day of the launch.

This is the quietly impressive part. The cheap model is not the stale model: Haiku 5.5 has the same June 2026 cutoff as Anthropic's most expensive models, and eleven months more than the Haiku it replaces. Its chat context window went from Haiku 4.5's 200K tokens to 1M.

Two caveats worth having. Anthropic's context-window article is titled and scoped "on paid Claude plans," and states no number for the free plan, so we are not printing one. And the context window differs by product: the same article gives Haiku 5.5 1M tokens in Claude Code and 500K in Claude Cowork.

It Watermarks What It Writes

Text written by Haiku 5.5 carries an invisible watermark from day one, which is now normal for a new Claude model rather than notable. Anthropic's help centre article on marking AI-generated content added a Haiku 5.5 row ticked in all three columns: text watermarks on Anthropic's own surfaces, text watermarks through AWS, Google Cloud and Microsoft Foundry, and Content Credentials on files.

That table now ticks nine of its sixteen models for text, and the newest additions reach the free plan: Haiku 5.5, Sonnet 5.5 and, in the revision stamped October 8, 2026, Sonnet 5. Whichever of the three a free user selects, the text is marked. The mark adds nothing to your writing, costs nothing, and Anthropic says it "carries no identifying information and can't be traced to a specific person, organization, or chat." Our explainer on Claude's text watermark covers what one can and cannot prove.

Effort Is Yours, Thinking Is Not

Haiku 5.5 is the first Haiku you can turn up and down, and the dial is in the same place as the model name. Anthropic's model settings article, revised on launch day, now lists Haiku 5.5 among the models with an effort selector: "The effort selector is available for Claude Haiku 5.5, Sonnet 5.5, Opus 5.5, Fable 5.1, Opus 5, Sonnet 5, Fable 5, Opus 4.7, Opus 4.6, and Sonnet 4.6."

This is the setting a free user should actually care about, because effort is what decides how fast you hit the wall. Anthropic's own description: Low and Medium "work well for routine tasks and stretch your usage further," High "offers the best overall balance of quality and speed," and Max is the most thorough and slowest. Anthropic no longer names one level as the app default; it says each model has its own recommended level "marked as 'Default' in the menu," so look rather than assume.

What you cannot do is switch thinking off. The same article says thinking "cannot be turned off in Claude when using Claude Haiku 5.5, Claude Sonnet 5.5, Claude Opus 5.5, Claude Fable 5.1, or Claude Opus 5." On the API it is different, and the wording is precise: for Haiku 5.5 "thinking can be turned off at effort levels high and below, but attempting to disable thinking at xhigh or max effort returns an error."

No Haiku-specific usage limit has been published. Anthropic's usage and length limits article was revised on October 8, 2026 and does not mention Haiku at all, so there is no stated allowance for it, better or worse than any other model.

The Part That Puts Money Back: Max and Team API Credits

If you pay for Max or Team, this release hands you a monthly credit that you have to go and claim yourself, and it expires if you do not spend it. Anthropic announced it alongside Haiku 5.5 and published the detail in a help centre article stamped October 7, 2026.

Anthropic's exact wording on the Team pool: credits "for all seats are pooled into one monthly balance, capped at $500," and its own worked example is "a team with three Standard seats and two Premium seats receives $260 a month."

The conditions are where people will trip:

• You claim it, it is not applied automatically. On claude.ai in a web browser, go to Settings > Billing as a Max subscriber or Organization settings > Billing as a Team owner, then "Link organization" in the API credits section.
• Seven days on the plan first. Anthropic: "New subscribers can claim once they've been on an eligible plan for seven days."
• One Console organisation, and you cannot change it yourself. "You can link only one Console organization... You can't change the linked organization yourself, so choose the one you plan to build in."
• It does not roll over. "Unused credits expire at the end of each billing cycle."
• It is not for Claude Code. The credits cover the API, the Console Playground, Claude Managed Agents and the Agent SDK. Anthropic is explicit that they do not apply to "Interactive Claude Code in the terminal, IDE, desktop, or web," nor to extra usage in Claude or Claude Cowork, nor to Claude on Bedrock, Vertex AI or Microsoft Foundry.
• It does not change your chat limits. "The credits don't change your usage limits in Claude, Claude Code, or Claude Cowork."

So: genuinely free money if you were ever going to call the API, and genuinely useless if you were not. It does not make your Claude subscription go further.

What It Costs Developers

Much less than Haiku 4.5, and the pricing has a cliff in it at 100,000 tokens. Anthropic's published table:

Anthropic's footnote does the arithmetic honestly, including the part that works against it: Haiku 5.5 is "priced 90% lower than Claude Haiku 4.5 for requests up to 100,000 tokens, and 50% lower for requests over," and it has "an updated tokenizer... which means it uses slightly more tokens per task." Its net claim is around 75% cheaper on average.

One other price moved in the same post, and it is not about Haiku. Anthropic halved Sonnet 5.5's cache reads from $0.20 to $0.10 per million tokens, which it estimates cuts the cost of Sonnet 5.5 on most agentic work by about 20%. That reaches you only if you pay per token.

Should You Change Anything?

On free Claude: open your model menu and know it is there. You now have three models to choose between, and the useful mental split is simple. Haiku 5.5 for speed on something simple, Sonnet 5.5 for everything else. Do not switch to Haiku as a general upgrade; it is not one, and Anthropic does not claim it is.

On Pro: nothing to do, and nothing to claim. Haiku 5.5 is included, no limit changed, and Pro is explicitly not eligible for the new API credit. Anthropic answers that question itself in the article's FAQ: "I'm on a Pro plan. Can I get this credit? No."

On Max or Team: go and claim the credit. It is $100 to $500 a month depending on your plan, it expires monthly if unspent, and the claim is a one-time link to a single Console organisation you cannot change later. Do it deliberately.

Do not start paying because of this release. Everything consumer-facing in it is already on the free plan. If you are weighing a subscription for other reasons, which AI chatbot is worth paying for works through it, and the best free AI chatbot in 2026 covers how Claude's free tier compares with ChatGPT and Gemini.

If you build with the API, re-run your numbers at the 100K boundary. The 10x step between the two price bands is bigger than the gap between Haiku 5.5 and some larger models, and Anthropic says 90% of requests to the previous Haiku fell under 100,000 tokens.

The bottom line: Claude's cheapest model is free on every plan, knows as much as the expensive ones, and is the fastest Claude yet, and it still should not be your default. Anthropic built it for volume and says so. The two things to carry away are that you have to select it yourself, and that if you pay for Max or Team there is a monthly API credit sitting unclaimed in your billing settings.

Sources

• Anthropic, Introducing Claude Haiku 5.5, page dated October 7, 2026
• Anthropic, Claude Haiku model page, "Availability and pricing," read October 12, 2026
• Anthropic, Release notes, help centre, entry dated October 7, 2026: "Claude Haiku 5.5 launch," lastUpdatedDate 2026-10-07T18:03:17Z
• Anthropic, Monthly API credits for Max and Team plans, lastUpdatedDate 2026-10-07T22:53:26Z, read October 12, 2026
• Anthropic, How up to date is Claude's training data, lastUpdatedDate 2026-10-07T17:55:41Z
• Anthropic, How large is the context window on paid Claude plans, lastUpdatedDate 2026-10-07T21:19:50Z
• Anthropic, Change the model, effort, and thinking settings, lastUpdatedDate 2026-10-07T17:55:42Z
• Anthropic, How Claude marks AI-generated content, lastUpdatedDate 2026-10-08T23:39:20Z
• Anthropic, How do usage and length limits work, lastUpdatedDate 2026-10-08T19:02:45Z, which does not mention Haiku
• Anthropic, claude.com/pricing plan comparison matrix, read October 12, 2026