# Harvey's Tenet model doubles legal task completion on benchmark

> Post‑training a Kimi K3 model with Harvey improves all‑pass rates on the Legal Agent Benchmark and keeps per‑task cost flat.

Oossa · 2026-10-23 · https://oossa.com/en/harvey-s-tenet-model-doubles-legal-task-completion-on-benchmark

Harvey and Fireworks AI have finished post‑training a Kimi K3 base model, creating a new model called Tenet. Tenet scores 19.7% all‑pass on the Legal Agent Benchmark, about twice the number of tasks completed and more than double the all‑pass rate of the original Kimi K3 (10.8%). The cost per task stays roughly the same at $5.92 versus $5.62 for the base model. The gains also show up on other legal agent tests the model never saw during training.

## The facts

- Tenet all‑pass on LAB: 19.7% vs. Kimi K3 10.8%
- Per‑task cost: $5.92 for Tenet, $5.62 for Kimi K3

## Why it matters

Law firms can run more complex AI‑assisted legal tasks without higher per‑task fees.

## Sources & references

1. [Post-training Kimi K3 with Harvey for long-horizon legal work](https://fireworks.ai/blog/post-training-kimi-k3-with-harvey-for-long-horizon-legal-work/) – Fireworks AI

Last updated: 2026-10-23