Claude Haiku 5.5 is now available in ThinkReview
Claude Haiku 5.5 is now available in ThinkReview
We're excited to announce that Claude Haiku 5.5 is now available in ThinkReview. On October 7, Anthropic released Claude Haiku 5.5, and the announcement opens with this:
Introducing Claude Haiku 5.5: the cheapest, fastest, and most capable small model we've ever released.
You can use it today for AI-powered code reviews on GitHub, GitLab, Azure DevOps, and Bitbucket. For the side-by-side with the model it replaces, see Claude Haiku 5.5 vs Haiku 4.5.
What Anthropic shipped
Haiku 5.5 is the small model in the Claude 5.5 family, built for high-volume work that has to stay cheap and fast. Anthropic names the jobs: summaries, compaction, classification, database queries, and subagent tasks next to Opus 5.5 and Sonnet 5.5. Developers call it as claude-haiku-5-5.
The specs that change a review:
- About 75% less to run than Haiku 4.5, on Anthropic's average. List price for prompts up to 100,000 tokens is $0.10 per million input tokens and $0.50 per million output tokens, down from Haiku 4.5's $1 / $5.
- A 1 million token context window, up from 200,000 on Haiku 4.5, with max output doubled from 64,000 tokens to 128,000.
- Adaptive thinking, the first time a Haiku ships with an adjustable effort setting. Thinking is on by default. The API default effort is
medium. - The fastest Claude at standard speed. Anthropic's footnote: it still runs slower than Opus models in Fast Mode.
- Knowledge cutoff of June 2026, and the same newer tokenizer as Sonnet 5.5 and Opus 5.5. The same text counts as about 30% more tokens than it did on Haiku 4.5.
The same announcement also cuts cache reads on Sonnet 5.5 from $0.20 to $0.10 per million tokens. Anthropic says that takes about 20% off the cost of most agentic work on Sonnet 5.5.
The scores that moved
Anthropic's launch table is the short version of the jump from Haiku 4.5. Sonnet 5.5 is the reference column, included so the small model has a ceiling next to it.
| Evaluation | Haiku 5.5 | Haiku 4.5 | Sonnet 5.5 |
|---|---|---|---|
| GDPval-AA v2.1 | 1620 | 735 | 1840 |
| Terminal-Bench 4.0 | 39.2% | 0.0% | 70.6% |
| FrontierCode 1.1 (Main) | 46.4% | — | 52.1% (Xhigh) |
| OSWorld 2.1 (offline subset) | 72.4% | 15.7% | 83.9% |
Source: Introducing Claude Haiku 5.5, October 7, 2026. Anthropic did not publish a Haiku 4.5 score on FrontierCode 1.1. Sonnet 5.5's FrontierCode figure is at Xhigh effort.
GDPval-AA is real professional work across 44 occupations. Haiku 5.5 scores 1620, against 735 for Haiku 4.5 and 1840 for Sonnet 5.5. It also passes GPT-6 Luna's 1437 on that row.
FrontierCode asks whether an agent's code change would be merged. Haiku 5.5 scores 46.4%. Sonnet 5.5 at Xhigh effort scores 52.1%. GPT-6 Luna scores 42.4% on the same row.
Terminal-Bench 4.0 is the number Anthropic uses to draw the line. Haiku 5.5 scores 39.2%. Sonnet 5.5 scores 70.6%. Their reading, from the announcement: Sonnet 5.5 and Opus 5.5 remain the better choice for complex agentic coding. Haiku 5.5 is for narrowly scoped tasks that used to be too expensive to run on every change.
Early customers reported the same shape. Aaron Vinh, Staff Software Engineer at Asana, measured over a 30% drop in task-completion latency and up to 2.5× faster inference per agent turn on bug triage, project setup, and portfolio search. Yashodha Bhavnani, VP of AI Products at Box, said Haiku 5.5 scored 11 points higher than Haiku 4.5 at about half the latency, on analytical work that runs at scale, including weekly recurring reviews.
What this means for code review
Most pull requests are the workload this price is for: a defined change, a comment a human will act on, and a queue long enough that Sonnet rates on every diff get expensive.
- The everyday review can stay on Haiku. At $0.10 / $0.50 under 100,000 tokens, reviewing the small and medium changes in a queue is a different bill from Haiku 4.5's $1 / $5. Cache reads on that band are $0.01 per million tokens, against $0.10 on Haiku 4.5, which matters when the same repository context is reused across reviews.
- The new 1M window holds the diff and the files around it. Haiku 4.5 stopped at 200,000 tokens. A wide change plus repository-level context can fill that before the model writes the comment. One million tokens keeps the surrounding code in the same pass. Prompts over 100,000 tokens step up to $0.50 / $2.50, which is still half of Haiku 4.5.
- Adaptive thinking spends effort where the hunk needs it. The model decides how much to think. A trivial patch can finish on a short pass. A change that has to follow a call across files gets a longer one, on Haiku rates. Effort is the dial Anthropic added. The default is medium.
- Keep Sonnet for the open-ended review. Terminal-Bench is the tell. When the question is a multi-step trace through a design, Sonnet 5.5 is the model Anthropic points at. Pair either one with repository context when the real bug lives outside the hunk.
Vendor tables are not your team's review bar. The honest comparison is still the same diff, the same instructions, and two models.
How to use Claude Haiku 5.5 in ThinkReview
- Open ThinkReview settings ��� Click the extension icon and go to Settings or Model selection.
- Choose Claude Haiku 5.5 — Select it from the model dropdown for your reviews.
- Run a review — Open any pull request or merge request and start ThinkReview with your chosen model.
Claude Haiku 5.5 sits on Lite (and plans that include Lite models), the same tier as Haiku 4.5. Manage your catalog in Model Selection on the ThinkReview portal.
Where it works
Same as always: GitHub, GitLab, Azure DevOps, and Bitbucket Cloud. One extension, one workflow — now with Anthropic's Haiku 5.5.
If you try Claude Haiku 5.5 on real pull requests and have feedback, we'd love to hear it — open an issue on GitHub or reach out via thinkreview.dev.
Ready for cheaper Haiku-class reviews? Install ThinkReview or manage your models in the portal.
Quotes, prices, and benchmark figures reference Anthropic's Claude Haiku 5.5 announcement, October 7, 2026, and the Haiku 5.5 model overview.