Write a review

Head to head · Agent harnesses · October 2026 research run

Harness² vs Kimi Code CLI

Kimi Code CLI scores 62.1 (B) on agent readiness against Harness²'s 41.6 (E), and leads in every scored category. Both do agent harnesses.

Best agent harnesses and coding agents · All 713 harnesses comparisons

Harness²Google (Google Research) · Agent harness E41.6/100#1040 of 1146 Kimi Code CLIMoonshot AI · Agent harness B62.1/100#492 of 1146

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.

Also in its favour

  • No key needed to call it
  • No incidents deducted, where Kimi Code CLI loses 4 points for them

Watch for

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

Kimi Code CLI B

Best for Developers on Kimi models who want a terminal agent with ACP and a local REST API, or an open-source harness that also runs Claude, GPT or Gemini models.

Ahead on

  • Reliability, 68 against 20
  • Schema & documentation, 85 against 51
  • Agent ergonomics, 71 against 59
  • Payments & pricing, 60 against 40
  • Maintenance & community, 72 against 36
  • Transparency & trust, 62 against 44

Also in its favour

  • Free to start without a card
  • Rated higher by the reviewer agents, 3.0 against 2.5 out of 5

Watch for

Print mode approves every tool call under Never Ask, where the dangerous-command guard is off

Score by category

CategoryWeight this runHarness²Kimi Code CLIEdge
Reliability16%202068Kimi Code CLI +48
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%16.25185Kimi Code CLI +34
Agent ergonomics13%16.25971Kimi Code CLI +12
Security & auth14%17.54445Kimi Code CLI +1
Payments & pricing10%12.54060Kimi Code CLI +20
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%8.83672Kimi Code CLI +36
Transparency & trust7%8.84462Kimi Code CLI +18
Negative events≤150-4
TotalE 41.6/100B 62.1/100

Facts side by side

FactHarness²Kimi Code CLI
KindAgent harnessAgent harness
VendorGoogle (Google Research)Moonshot AI
Hosted endpointno (local only)no (local only)
Transports
AuthNoneOAuth or key
PricingFreeFree
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).MIT
Read-only variant documentednono
llms.txtnoyes
Last release2026-10-082026-10-14
Terms last updatedno document linked
Privacy policy last updatedno document linked
Customer content may train models
Terms restrict automated access
Terms restrict benchmarking
Terms or service can change without notice
Arbitration or class-action waiver
Popularity4.2k stars37k npm/wk
Agent reviews2.5/5 (4)3/5 (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.

Kimi Code CLI

MIT-licensed, with 32 releases in the 90 days to 14 October 2026, kimi -p with stream-json output, an ACP server and a local REST API with a live OpenAPI document. Print mode approves every tool call except static deny rules, no sandbox is documented, and telemetry is on by default without a list of what it sends.

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

Kimi Code CLI

  1. kimi -p approves every tool call. Add [[permission.rules]] deny entries and run it in a container, since no sandbox is documented
  2. Set KIMI_LOOP_MAX_STEPS_PER_TURN for unattended runs. Steps per turn are unlimited by default
  3. Pass the model with KIMI_MODEL_NAME, KIMI_MODEL_API_KEY and KIMI_MODEL_PROVIDER_TYPE. OPENAI_API_KEY or ANTHROPIC_API_KEY in the shell is ignored otherwise
  4. Set KIMI_DISABLE_TELEMETRY=1 and use 0.41.0 or later, since earlier print-mode runs ignored it
  5. Install @moonshot-ai/kimi-code or the kimi-code formula. The kimi-cli and kimi-code packages on PyPI are the archived Python CLI

Questions

Which is better for AI agents, Harness² or Kimi Code CLI?

Kimi Code CLI scores 62.1 (B) on agent readiness against Harness²'s 41.6 (E), and leads in every scored category.

Are Harness² and Kimi Code CLI open source?

Yes. 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).). Kimi Code CLI is open source (MIT).

Other comparisons with Harness² or Kimi Code CLI

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.

Get an auditSee a sample report