AIEvals

Benchmarks

DRACO (Anthropic harness)

Accuracy, completeness, presentation and citation quality on 100 complex research tasks drawn from user queries.

Deep researchReported by AnthropicVendor-reported onlyReported by Anthropic

How it is run

Section 8.8.2 uses web tools and a 980k-token budget. Claude Opus 4.6 judges the final report file with five grading runs; labelled effort curves are retained.

What it cannot tell you

The original paper uses a different judge and system prompt. Judge choice can shift scores by 10–25 points, so headline results are not directly comparable.

Why there is no independent table

No independent suite in this dataset runs DRACO (Anthropic harness). The numbers below are the ones model publishers report about their own models, which is a different kind of evidence: each publisher chooses its own harness and effort level, so a row is readable on its own and not against the other rows.

Read this with

Nothing else on this site shares this benchmark's family or category, so these are other evaluations the same publisher reports.

As reported by each publisher

Kept separate per publisher, because the settings are not shared.

Anthropic

Claude Haiku 5.5: system card capability evaluations · 7 October 2026

RankModelScore
3Claude Haiku 5.581.5%
4Claude Haiku 4.552.4%
2Claude Sonnet 5.587.0%
1Claude Opus 5.587.4%

Ranks compare reported scores in the same published row; ties share a rank.

DRACO (Anthropic harness). Figure 8.8.2.A labelled max scores. Haiku low/medium/high/xhigh: 64.3/72.4/77.8/80.5%. Opus 4.6 judge; 980k task budget, five grading runs per final report.

Section 8, pages 111–139 of the Haiku 5.5 system card, read 8 October 2026. Values come from exact prose, tables and labelled figures. Unless specified, Haiku uses adaptive thinking at max effort and five trials with evaluation-specific context up to 1M. Each row retains its subset, metric, effort and source notes; no score is assigned to an unspecified SWE-Bench Pro version. Terminal runs use safeguards without fallback and no internet egress. ProgramBench uses filtered tasks and hidden-test pass rate without a time limit. OSWorld partial and strict results use the official 82-task offline subset. CAD uses 0–1 voxel IoU. Healthcare raw and length-adjusted scores stay separate. Life-science results disable Haiku biology safeguards. Independently run figures quoted by Anthropic remain publisher reports here. Missing comparison values are not inferred from rounded charts.

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

Sources

Every page used for this benchmark, read on 17 October 2026.