AI DevelopmentMethodology5 min readPublished September 9, 2026

Terminal-Bench 4.0: When Agent Scores Need a New Test

Terminal-Bench 4.0 changes the test behind agent scores. Decide when results need rerunning, regrading or reuse before comparing coding agents for your team.

DA
Digital Applied Team
AI research and implementation
Editorial dateSeptember 9, 2026
ReviewedSeptember 12, 2026

Terminal-Bench 4.0 changes the conditions behind an agent score. Before using an older result to choose a coding agent, identify which tasks and settings changed, then decide whether the evidence requires a new run, a new grade or no substantive change. A model name and a percentage are not enough.

The official v4.0.0 release was published on August 26, 2026. It identifies removed tasks and changes to existing ones. The useful buying question is which of your comparison rows remain supported, rather than whether the new version makes an agent look better.

Key takeaways
  1. 01
    Match versions before comparing.Changed tasks or resources can alter a result without changing the model.
  2. 02
    Separate rerunning from regrading.A new verifier may use retained artifacts; a changed agent environment may require new work.
  3. 03
    Record actual settings.Defaults and prose summaries do not prove which overrides a submitted run used.

01Practical guideChoose between rerun, regrade and reuse

Treat an existing result as evidence produced by a specific experiment. Rerunning repeats the agent’s work. Regrading applies an appropriate verifier to retained outputs. Reuse carries a result forward when the relevant experiment has not materially changed. Each route needs a reason.

The benchmark maintainers describe these distinctions in their continuous-benchmark methodology . Apply the change classification to the affected tasks. A major dataset update does not by itself establish that every individual task needs the same treatment.

Digital Applied proposed method; source-backed distinctions explained in the accompanying text.
ChangeCandidate actionEvidence needed
Task input or execution environment changesRerun affected workExact new task, resources, agent settings and retained outputs.
Verifier changes without changing the agent taskRegrade if the artifacts are sufficientCompatible final artifacts, verifier version and both grading outcomes.
Documentation-only correctionReuse if behavior and grading are unchangedChange review confirming no material effect on the experiment.
Task removed or denominator changedRecompute the comparison on a stated setIncluded task manifest and scoring policy.
Change cannot be classifiedLeave the result unresolvedMore information from the benchmark or run owner.

02Practical guideInspect the executable configuration

A concrete example appears in the pinned v4.0.0 task for batched-eval-parity. Its README summary describes four hours and one CPU, while the changelog describes eight hours and two CPUs. The task configuration specifies an agent timeout of 28,800 seconds and two CPUs.

That is a documentation/configuration discrepancy observed by reading the pinned files, not a benchmark run. It shows why the source and revision matter. The configuration is more directly relevant to execution than a stale summary, but even it does not prove the settings of a particular submitted result. Overrides may still apply.

Ask for the actual run configuration: resource allocation, hard limits, timeout, concurrency, agent revision and retry policy. Keep the source of every field. If a provider reports its default rather than the run’s value, mark that difference instead of filling the gap with an assumption.

03Practical guideKeep resources in the comparison

Anthropic’s infrastructure study reports that resource settings changed results in its Terminal-Bench 2.0 experiments while the model, harness and tasks were held fixed. It distinguishes guaranteed resources from hard limits. This is vendor-run evidence about that experiment, not a universal correction factor for version 4.0.

A larger allowance can change what an agent has time to attempt, while a hard limit can determine whether a process survives. Record both when they matter. Do not infer that increasing every resource always improves the score or that one percentage adjustment makes unlike experiments comparable.

Our OSWorld and Terminal-Bench explainer covers what the benchmark families measure. This article adds the migration question: how to preserve a defensible comparison when the measurement system itself changes.

04Practical guideBuild a migration record instead of a mixed leaderboard

Give each old-to-new comparison a record with its own inclusion rules. If the new dataset removes tasks, a shared model label does not preserve the old denominator. A score is meaningful only with the task set and outcome accounting that produced it.

Digital Applied proposed method; source-backed distinctions explained in the accompanying text.
RecordIncludeReason
Dataset identityOld and target versions, immutable references and task manifestsMakes additions, removals and changed tasks visible.
Agent identityHarness revision, model identifier, reasoning and retry settingsDistinguishes a model change from a system change.
ExecutionResource allocations, hard limits, timeout and actual overridesPrevents unlike environments from appearing equivalent.
VerificationVerifier revision, available artifacts and migration actionExplains whether the result was rerun, regraded or reused.
OutcomeTask denominator, attempts, success, timeout and error handlingPreserves the publisher’s scoring policy.
ProvenanceRun owner, date, original record and measured costs if suppliedSeparates vendor reports from independent reproduction.

05Practical guideAsk for the evidence that changes the buying decision

Suppose a supplier shows an older result beside a version-4 score from another agent. Ask for the common task set, each run’s configuration and the policy used to carry results forward. If those are unavailable, treat the comparison as incomplete. You can still investigate both products without pretending their numbers are directly comparable.

The vendor benchmark reproducibility audit provides a disclosure baseline. The broader SWE-Bench and Terminal-Bench guide explains their differing tasks. Neither source should be used to manufacture a conversion between benchmark versions.

Avoid importing current leaderboard percentages into a historical comparison without a dated export. A present-day board can change after a submission correction or new run. Retain the original record that supported the decision, including unsuccessful attempts and missing fields.

06Practical guideKeep the migration narrow enough to audit

Start with the result you actually use to shortlist an agent. Classify its relevant changes and request only the missing evidence. If a new run is needed, freeze its task set and settings before comparing it with another system. If regrading is possible, retain the old artifacts and both verifier outputs.

No benchmark tasks were executed for this article, and no model ranking is inferred from the release. The migration table is a proposed editorial method. For help connecting evaluation evidence to a working application, our AI transformation service can scope an evaluation around the tasks that matter to your team.

Methodology

Evidence and scope

As-of date
September 12, 2026. September 9 is the editorial allocation; this research was reviewed later.
Method
Primary release documentation and pinned task files were inspected. The result-migration table is a proposed method for buyers; it is not a live leaderboard.
Limitations
No benchmark runs, model comparisons or cost measurements were performed. Current methodology pages provide context; release-specific claims use the pinned August 26 artifact.

07Next stepCompare experiments that support the same question.

Put it into practice

Compare experiments that support the same question.

When a benchmark changes, preserve the old evidence and explain the action required for each affected result. An unresolved comparison is more useful than a precise-looking ranking built from incompatible runs.

From AI output to accepted work

Make your next AI workflow reviewable.

Define the result, the evidence and the people responsible for acceptance.

Clear scopePractical evaluationAccountable delivery
Implementation

Build around the result you need

  • Choose a representative workflow
  • Define acceptance evidence
  • Review the delivered outcome
Questions and answers

Applying the method

Not necessarily. Inspect the changes and benchmark policy at task level. Some tasks may be unchanged while others require a new environment, new grading or removal from the comparison.