Independent newsroom

Thursday, October 8, 2026

The File

Every claim traces to a primary source. Uncertainty is labeled. Corrections are public.

Anthropic · Thursday, October 8, 2026

Anthropic releases Claude Haiku 5.5, a cheaper and faster small model

The File desk · Oct 8, 2026, 10:54 AM UTC

Status

Confirmed. Anthropic's launch page, dated Oct. 7, 2026, plus a system card dated the same day. Every cost and performance number below is Anthropic's. This desk did not check them independently. The system card calls Haiku 5.5 one of the company's non-frontier models.

  • Confirmed

    Anthropic released Claude Haiku 5.5 on Oct. 7. The model ID is claude-haiku-5-5. A system card dated Oct. 7 accompanies the launch.

  • Confirmed

    Anthropic says the model costs around 75 percent less to run than Haiku 4.5 on average. Those figures are the company's.

Anthropic released Claude Haiku 5.5 on Wednesday, the smallest and cheapest model in its new 5.5 family. The company says it is the most capable small model it has made, and that it costs around 75 percent less to run than Haiku 4.5, the model it replaces. It is built for quick, repetitive jobs, and for working as a helper under Anthropic's bigger models. The model ID is claude-haiku-5-5. A system card dated Oct. 7 calls it one of the company's non-frontier models. Sept. 23 and Sept. 29 stories on this site said Haiku 5.5 was coming.

Anthropic's launch page says Haiku 5.5 is “the cheapest, fastest, and most capable small model we've ever released.” On average, the company says, it now costs around 75 percent less to run than Haiku 4.5. Those are Anthropic's figures.

The model ID is claude-haiku-5-5. Anthropic's API notes say it has a context window of 1 million tokens and up to 128,000 output tokens, and that it is the first Haiku with an adjustable effort setting. For prompts up to 100,000 tokens, Anthropic lists $0.10 per million input tokens and $0.50 per million output tokens. Over 100,000 tokens, those prices are $0.50 and $2.50. Haiku 4.5 is listed at $1 and $5. Anthropic says Haiku 5.5 is priced 90 percent lower than Haiku 4.5 for shorter requests, which were about 90 percent of Haiku 4.5 traffic, and that it uses slightly more tokens per task. That is how the company gets to around 75 percent less on average.

On Anthropic's own benchmark table, Haiku 5.5 scored 39.2 percent on Terminal-Bench 4.0, a test of coding at a command line, against 0.0 percent for Haiku 4.5 and 70.6 percent for Sonnet 5.5. On an offline subset of OSWorld 2.1, a test of operating a computer, Anthropic reports 72.4 percent for Haiku 5.5 and 15.7 percent for Haiku 4.5. Anthropic says Sonnet 5.5 and Opus 5.5 remain better choices for complex coding work that runs on its own. All of those scores are Anthropic's.

The system card says tests under Anthropic's

, the company's internal rules for checking models for dangerous abilities before release, found Haiku 5.5 broadly less capable than Claude Opus 5, and that it does not cross any new thresholds in those rules. The card says the model's cyber safeguards are looser than those on other recent Anthropic models but tighter than Haiku 4.5's, and that they still block penetration testing. It also says the model over-refused more than any other model Anthropic tested in an automated audit. The card says the company expects system cards for its non-frontier models to be similarly condensed going forward. The knowledge cutoff is June 2026.

Anthropic says Haiku 5.5 is available on the Claude API, Amazon Bedrock, Claude Platform on Amazon Web Services, Google Cloud, and Microsoft Foundry. The same notes cut Sonnet 5.5 cache reads from $0.20 to $0.10 per million tokens, which Anthropic says is about 20 percent cheaper on most agent work. New monthly API credits this week are $100 for Max 5x, $200 for Max 20x, and up to $500 pooled for Team. The release notes do not say the model is on a free plan.

This site's Sept. 23 story on Claude Opus 5.5, and the Sept. 29 story on Claude Sonnet 5.5, both said Haiku 5.5 would follow in the coming weeks. Wednesday was the launch.

What is still unknown or disputed

Primary sources

Every claim in this story is drawn from the documents below. If a fetch failed, that is recorded on the card.

  1. Source 1

    Introducing Claude Haiku 5.5

    Anthropic · October 7, 2026

    Introducing Claude Haiku 5.5: the cheapest, fastest, and most capable small model we've ever released.

    https://www.anthropic.com/claude-haiku-5-5

  2. Source 2

    System Card: Claude Haiku 5.5

    Anthropic · October 7, 2026

    Our (RSP) evaluations found that Claude Haiku 5.5 is broadly less capable than Claude Opus 5 across domains and does not cross any new RSP thresholds.

    https://www.anthropic.com/claude-haiku-5-5-system-card

  3. Source 3

    Claude API release notes, Oct. 7, 2026

    Anthropic · October 7, 2026

    We've launched Claude Haiku 5.5 (claude-haiku-5-5), our most capable model tuned for high-volume and latency-sensitive work. It has a 1M token context window, 128k max output tokens, and adaptive thinking with the effort parameter.

    https://docs.claude.com/en/release-notes/api

  4. Source 4

    Anthropic ships Claude Opus 5.5

    The File · September 23, 2026

    Sonnet 5.5 and Haiku 5.5 are said to follow in coming weeks.

    /story/anthropic-claude-opus-5-5-sept22

  5. Source 5

    Anthropic ships Claude Sonnet 5.5

    The File · September 29, 2026

    Anthropic says Claude Haiku 5.5 will join the family in the coming weeks.

    /story/anthropic-claude-sonnet-5-5-sept28