docs: Terminal-Bench 2.1 Maka vs Kimi Code v11 results (#1217)
main ← docs/tb21-maka-vs-kimi-code-v11-report
已合并 03:36AM - 19 Jul 26 UTC
## Summary
Docs-only: English write-up of Terminal-Bench 2.1 harness A/B run `k…3-maka-vs-kimi-code-tbench-2.1-full-v11` (Maka vs Kimi Code, both **k3** / **max** thinking), plus a redaction-minimal accepted-outcome CSV.
**Full report:** [`docs/eval/terminal-bench-2.1-maka-vs-kimi-code-v11.md`](https://github.com/maka-agent/maka-agent/blob/docs/tb21-maka-vs-kimi-code-v11-report/docs/eval/terminal-bench-2.1-maka-vs-kimi-code-v11.md)
### Results
| | Maka | Kimi Code |
| --- | ---: | ---: |
| **Pass@1 (accepted)** | **62/89 (69.7%)** | **53/89 (59.6%)** |
| Mechanical first verifier | 62/89 | 50/89 (56.2%) |
| Deadline-killed | 28/89 | 33/89 |
| Pass rate, finished in time | **95.1%** | **85.7%** |
| Pass rate when killed | 14.3% | 15.2% |
| List-price cost (88 pairs) | $52.72 | $38.40 |
Official reference: **88.3%** for Kimi K3 + KimiCode harness ([tech blog](https://www.kimi.com/blog/kimi-k3), accessed 2026-07-19).
Kimi Code's accepted 53/89 is operator-adjusted (mechanical first verifier = **50/89**). Three score-changing recoveries: `prove-plus-comm`, `hf-model-inference`, `pypi-server`. Run admissions: 203 physical → 178 accepted (25 recovery). Single-repetition; descriptive, not statistical.
### Absolute scores are throughput-limited
Decode was ~34–37 output tok/s with ~14 s median first-token latency against task-native agent budgets. About a third of tasks per arm hit the deadline mid-generation and rarely passed (14–15%).
Finished-in-time pass rates (95.1% / 85.7%; **95.9% / 89.8%** on the 49 tasks both arms finished) sit around the official 88.3%. Under healthy throughput, Maka would plausibly reach or exceed that reference and Kimi Code land near it — subsets are self-selected, so a re-run is decisive.
Official frontier is already compressed (~88.8% / 88.3%). Unconstrained subsets pass at 85.7–95.9%. **TB 2.1 is close to saturated at the frontier; harder benchmarks will carry more signal.**
### Relative gap is harness asymmetry
Thinking effort, thinking retention, and wall-clock are aligned (same degraded network path). The +10.1 pp (Maka) persists among finished tasks (95.1% vs 85.7%). Differences below are **listed, not ranked** as score causes.
| # | Difference | Notes |
| --- | --- | --- |
| 1 | **Context-budget tool-result prune (Maka only)** | 2048-token threshold ON; Kimi Code has no equivalent policy. Live 89/89; diagnostics 21/57 completed: 576 rewrites, ~1.87M est. tokens not re-sent (lower bound; aborted cells unmetered). Score impact not estimated (no prune-off arm). |
| 2 | System prompt | Empty vs ~20KB product main-agent template |
| 3 | Tool surface | Narrow foreground Bash vs full product CLI |
| 4 | Completion contract | Task+schemas vs product “tell the user you’re done” norms |
Background Bash is not the exclusive-win story (2/12 Maka-only tasks used `run_in_background` on Kimi). Heavy-task self-check is in the build but was **off** (zero tool/gate events). Recorded cost is $0 (account-plan); list-price uses official K3 API rates.
## Verification
- Load-bearing numbers recomputed from the frozen run: `controller/results.jsonl{,.attempts.jsonl}`, trial `result.json`, Kimi `provider-request-telemetry.json`, Maka `maka-cell-output.json` / runtime streams, `harness-ab-report.json`, task `task.toml` budgets
- Official 88.3% cross-checked via secondary sources quoting the tech blog table
- CSV 89/89 match to accepted outcomes; SHA-256 integrity table in the doc
- External review (Codex + Claude Fable); P0–P2 findings resolved
- Docs-only; no code, tests, or raw eval artifacts; `git diff --check` clean