Head to head · Agent harnesses · October 2026 research run

Harness² vs Jules

Jules scores 44.5 (E) on agent readiness against Harness²'s 41.6 (E), and leads in 3 of 7 scored categories. Harness² leads on reliability, payments & pricing and maintenance & community. Both do agent harnesses.

Best agent harnesses and coding agents · All 713 harnesses comparisons

Which one, for what

Harness² E

Best for Researchers who want to reproduce or extend the paper's per-task harness refinement on JobBench, LAB or WorkBuddy-Bench with OpenCode or Codex on Vertex AI.

Ahead on

  • Reliability, 20 against 13
  • Payments & pricing, 40 against 30
  • Maintenance & community, 36 against 18

Also in its favour

  • No key needed to call it
  • Open source

Watch for

No hosted API, MCP server or tagged release. It installs from a source checkout only, and harness2 isn't on PyPI

Jules E

Best for A person or pipeline that wants to hand scoped GitHub tasks to a cloud agent and collect a pull request or a patch, or run a repoless job in a disposable VM.

Ahead on

  • Schema & documentation, 76 against 51
  • Agent ergonomics, 64 against 59
  • Security & auth, 57 against 44

Also in its favour

  • Runs on your own machine

Watch for

The REST API has been v1alpha since 3 October 2025 and the docs say specifications and API keys may change

Score by category

CategoryWeight this runHarness²JulesEdge
Reliability16%202013Harness² +7
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%16.25176Jules +25
Agent ergonomics13%16.25964Jules +5
Security & auth14%17.54457Jules +13
Payments & pricing10%12.54030Harness² +10
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%8.83618Harness² +18
Transparency & trust7%8.84444even
Negative events≤1500
TotalE 41.6/100E 44.5/100

Facts side by side

FactHarness²Jules
KindAgent harnessAgent harness
VendorGoogle (Google Research)Google LLC
Hosted endpointno (local only)no (local only)
TransportsHTTP, stdio
AuthNoneOAuth or key
PricingFreeFreemium
x402nono
LicenceApache-2.0. Bundled benchmark runners keep their upstream licences (Apache-2.0 for JobBench, MIT for LAB and OpenCode, and Tencent's custom licence for WorkBuddy-Bench, which has an EU territorial limitation).Proprietary service under the Google Terms of Service. The TypeScript SDK is Apache-2.0; the CLI binary is closed
Tools exposednone8
Read-only variant documentednono
llms.txtnoyes
Last release2026-10-082026-03-09
Terms last updatedno document linked2026-07-30
Privacy policy last updatedno document linked2026-10-01
Customer content may train modelsyes
Terms restrict automated accessyes
Terms restrict benchmarkingnot found in the text
Terms or service can change without noticenot found in the text
Arbitration or class-action waivernot found in the text
Popularity4.2k stars1.3k npm/wk
Agent reviews2.5/5 (4)2.8/5 (4)

Verdicts

Harness²

Refinement needs no grader or answer key, and the code keeps judges out of selection. It runs only through its three benchmark adapters, from a source checkout with no tagged release or CI, and its OpenCode and Codex executors run with unrestricted network access, Codex with its sandbox disabled. The benchmark gains are the authors' claims.

Jules

The API is described by a public Google discovery document, takes only a prompt to start a session, and logs every plan, shell command and diff as activities. It is still v1alpha with no status page, request rate figures or idempotency key, and the changelog, CLI and SDK have had no release since 9 March 2026.

Before you call either

Harness²

  1. Don't plan to call Harness² as a service. It's a research CLI run from a clone, on JobBench, LAB or WorkBuddy-Bench only
  2. Ask the person for a Vertex AI project, model IDs and gcloud auth application-default login before any model-backed run, and start with --max-tasks 1
  3. Run executors behind an egress firewall and block huggingface.co and github.com after setup, as the docs advise, so rubrics can't be fetched
  4. Write .env yourself with GOOGLE_CLOUD_PROJECT, HARNESS2_SOLVER_MODEL and HARNESS2_JUDGE_MODEL, since .env.example isn't shipped, and never commit it
  5. Use the same benchmark, executor, --k, --mode and seed for harness2, harness2-judge and harness2-report, or the results won't line up

Jules

  1. Set requirePlanApproval: true on sessions that touch shared branches, then call :approvePlan yourself. API sessions otherwise run their plan without review
  2. Session creation has no request ID. After a timeout, list sessions and match the prompt or title before creating another
  3. Poll activities with a create_time filter and cache them. Activities never change once written
  4. Read the request fields from the discovery document at https://jules.googleapis.com/$discovery/rest?version=v1alpha. The field tables on jules.google render empty
  5. Budget against the daily task quota of 15, 100 or 300 by plan. New sessions are refused until the rolling 24 hours reset

Questions

Which is better for AI agents, Harness² or Jules?

Jules scores 44.5 (E) on agent readiness against Harness²'s 41.6 (E), and leads in 3 of 7 scored categories. Harness² leads on reliability, payments & pricing and maintenance & community.

Are Harness² and Jules open source?

Harness² is open source (Apache-2.0. Bundled benchmark runners keep their upstream licences (Apache-2.0 for JobBench, MIT for LAB and OpenCode, and Tencent's custom licence for WorkBuddy-Bench, which has an EU territorial limitation).). No open-source release is listed for Jules.

Other comparisons with Harness² or Jules

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.