AIEvals

Benchmarks

ProgramBench (hidden test pass rate)

Reconstructing real programs from their compiled binary and documentation, measured by behavioral hidden-test pass rate.

Long-context codingReported by AnthropicVendor-reported onlyReported by Anthropic

How it is run

Section 8.7.1 excludes 34 flaky tasks, leaving 166. It uses mini-swe-agent without the upstream six-hour limit and grades only tests passed by the reference binary.

What it cannot tell you

This partial hidden-test pass rate is not Vals fully resolved or almost-solved accuracy. Modified filtering and time limits make it a separate comparison.

Why there is no independent table

No independent suite in this dataset runs ProgramBench (hidden test pass rate). 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

Other versions of the same evaluation. Versions are not comparable: the task set changes between them.

  • ProgramBenchSame family. Whether an agent can produce a program that fully satisfies a specification, scored on complete resolution rather than partial credit.

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
2Claude Haiku 5.582.0%
3Claude Sonnet 5.579.7%
1Claude Opus 5.591.2%

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

ProgramBench (hidden test pass rate). §8.7.1: 166 filtered reconstruction tasks; mini-swe-agent without the six-hour limit. Hidden tests, not fully resolved task accuracy.

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.