DOOMBENCH.COM
Anthropic 🇺🇸 · Claude Haiku

Claude Haiku 5.5

Claude Haiku 5.5 is Anthropic's small hosted model, released as claude-haiku-5-5 with adjustable effort, a one-million-token context window and broad API/cloud availability. Provider evaluations show substantial computer-use, knowledge-work and coding gains over Haiku 4.5. The system card describes improved alignment but residual controlled-test failures.

DOOM SCORE78.2out of 100model risk profile, not the overall index
0 comments · 0 votesOpen discussion

Public discussion is readable by everyone. Sign in to comment, reply, or vote.

No comments yet. Start the discussion.

Send detailed or private feedback →
CURRENT ASSESSMENT · REVISION 1

Why this model scores 78.2

Capability 86: provider OSWorld offline 72.4%, GDPval 1620 and Terminal-Bench 39.2% improve on Haiku 4.5 but remain below larger frontiers. Autonomy 84: multi-step browser/computer use and coding subagents, not demonstrated frontier self-improvement. Deployment 100: dated API and AWS/Google/Azure availability, about 75% lower average task cost; commercial terms restrict duplication/reverse engineering. Misuse 70: safeguard-disabled cyber tests improve over 4.5, with limited high-end performance and deployed bio/cyber restrictions. Control difficulty 50: stronger agency and 4% staged sandbox violations, balanced by major injection/covert-action improvements; leaked-answer nondisclosure regresses. All safety failures are controlled tests, not external escapes. Exact model ID and access verified in October7 release notes; scores are editorial dimensions, not measured probabilities.

Capability86
Autonomy84
Deployment100
Misuse potential70
Control difficulty50
WATCH AND EXPLORE

Reviewed videos from Artificial Intelligence Videos. These links add context and never change a DoomBench score.

Thumbnail for Finally a good small model
41:10

Finally a good small model

Theo Browne tests Haiku 5.5, compares task costs and benchmarks, and explains why small models suit bounded subtasks more than full orchestration.

Theo Browne directly reviews Claude Haiku 5.5 as a bounded coding-agent worker, showing verification, API-limit and orchestration failures alongside its cost trade-offs.
Thumbnail for Haiku costs, interactive AI and SynthID limits
24:33

Haiku costs, interactive AI and SynthID limits

Nick Saraev and Jack Roberts discuss cheaper AI models, interactive ChatGPT explanations and the limits of detecting images with SynthID watermarks.

The reviewed page explicitly discusses Claude Haiku 5.5 capabilities and delegated workloads; its conflicting pricing units are not treated as verified comparisons.
Thumbnail for Bijan Bowen Tests Claude Haiku 5.5 on Games and Interfaces
31:27

Bijan Bowen Tests Claude Haiku 5.5 on Games and Interfaces

Bijan Bowen tests Claude Haiku 5.5 on browser interfaces and games, showing rendering defects, iterative fixes and his reported subscription usage.

Bijan Bowen directly tests Claude Haiku 5.5 on browser interfaces and games, showing rendering defects, repeated fixes and task limits alongside working output.
MODEL-ATTRIBUTED EVIDENCE

News tied to Claude Haiku 5.5

The model score of 78.2 rates this model's risk profile. The overall Doom Index of 67.9 measures the complete temporally weighted evidence record. These values answer different questions.

NET MODEL-ATTRIBUTED INDEX CONTRIBUTION+0.37

Each article's current Doom Index contribution is divided equally among the exact models named on that article. This prevents multi-model evidence from being claimed in full on several model pages. Model risk scores use a bounded temporal offset around their technical profile, but never feed back into the overall index.

Deployment TOWARD

Anthropic releases Haiku 5.5 with cheaper computer-use and coding agents

Anthropic released Claude Haiku 5.5 across its API, Amazon Web Services, Google Cloud and Microsoft Azure. It reports stronger computer-use and coding results than Haiku 4.5 and about 75% lower average task cost, after accounting for token usage. Broader inexpensive agent access is established; benchmark gains are provider-reported, not proof of job displacement.

Full item contribution
+0.37
Claude Haiku 5.5 equal share
+0.37
Read assessment →
AUDIT TRAIL

Model score history

  1. R1
    Doom Score 78.2

    Create the verified Haiku 5.5 version with dated access evidence and separate capability, autonomy, deployment, misuse and control assessments.

    08 Oct 2026

Share this page

DoomBench social sharing card for Claude Haiku 5.5.

  1. Claude Haiku 5.5 by Anthropic has a DoomBench model risk score of 78.2 out of 100, based on five transparent version-specific dimensions rather than the overall index.

  2. Claude Haiku 5.5's highest current DoomBench dimension is deployment at 100.0 out of 100; the profile publishes every component score and its editorial rationale.

  3. DoomBench links 1 source-backed evidence item to Claude Haiku 5.5, while keeping the model's risk profile separate from each item's contribution to the live Doom Index.

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