Harness² vs MiniMax Code

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.

MiniMax Code

An MIT-licensed terminal coding agent with a headless mode that returns versioned JSON, documented exit codes and an ACP server, and telemetry off by default. It's pre-1.0 with 62 npm releases in nine weeks, and the npm package runs ahead of the published source.

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

MiniMax Code

1. Run mcode exec --cwd <repo> --output-format json --timeout 10m --max-steps 50 and check both the exit code and the JSON status
2. Keep --permission full and off for a container or throwaway workspace. The default smart fails when it would need a person
3. Set MCODE_PROVIDER_API_KEY and add a provider for BYOK. M Plan models need a browser sign-in with mcode login --region global
4. Set MINIMAX_DATA_DIR per CI runner. Keys and sessions live in ~/.minimax by default
5. Pin a version. npm published 62 releases in nine weeks and the public source trails it