Switch models on evidence your counsel can check

Klu reruns your contract and compliance evals the day a new model ships. A cheaper model can't win by skipping a clause that should have gone to counsel.

Vendor contract reviewMuse Spark 1.3 rerun
Switch to Muse Spark 1.3. It sent every clause counsel had to see. Opus 5.5 High missed 3.
Escalations missed, Muse Spark 1.30 of 40▼ best
Escalations missed, Opus 5.5 High3 of 40▲ +3
Cost per contract, Opus 5.5 High$0.92→ today
Cost per contract, Muse Spark 1.3$0.39▼ −58%
Scored on 120 vendor contracts against your playbook and counsel's redlines. 40 contain a clause that must go to counsel.

A missed escalation costs more than a false flag.

Klu weights each run the way senior counsel reviews a junior's markup. Cases that must reach counsel are marked critical, and a model that misses one scores zero on that case.

01
Deviations caughtWhat breaks your playbook, not what it permits
35%
02
Right escalationUncapped exposure goes to counsel
25%
03
Quotes the sourceExact clause language, nothing invented
20%
04
Usable redlineYour fallback language, ready to send
15%
05
Cost per contractAt your monthly volume
5%

See the clause a model missed

Open any case to read the clause, the playbook rule, each model's answer, and why the grader scored it. Counsel can check the call in a minute instead of taking a number on trust.

Playbook review / Uncapped indemnity from usCritical
Clause 9.3

Customer shall defend, indemnify and hold harmless Vendor and its affiliates from any and all claims, losses and expenses arising out of or relating to Customer's use of the Services, and such obligations shall not be subject to any limitation of liability in this Agreement.

Playbook rule 2

Any indemnity we give is limited to our own breach or misuse and is capped. Uncapped exposure goes to counsel.

Muse Spark 1.3Verdict: Escalate. Uncapped indemnity from us. Limit it to our own breach or misuse and bring it under the cap.
96Passed
Opus 5.5 HighVerdict: Negotiate. Narrow "any and all claims" to claims arising from our breach of the Agreement.
0Critical case failed
Grader reason

The clause gives uncapped indemnity from us, so the verdict must be Escalate. The proposed redline is reasonable, but the contract never reaches counsel.

1 of 3 escalations Opus 5.5 High missed in this rerun

Four models beat every Claude on legal work

Harvey LAB is 120 legal tasks across 24 practice areas. A task passes only if every criterion does. The top seven sit within 4.3 points, so price decides. Klu reruns these models on your contracts.

Harvey LAB18 models · Oct 7, 2026
Tasks fully passed
1
Muse Spark 1.3 Xhigh$2 per 1M
95.5%
2
Kimi K3 Max$6 per 1M
94.6%
3
Qwen3.8 Max$3 per 1M
93.6%
4
Step 5 Preview$1.43 per 1M
93.4%
5
Fable 5.1 Xhigh$20 per 1M
93.3%
6
Sonnet 5.5 Max$4 per 1M
93.1%
7
Opus 5.5 Xhigh$8 per 1M
91.2%

Where legal and compliance teams use Klu

Each one starts with a default way to score it. Change it if your reviewers judge it differently.

Misses cost more

Playbook review

Vendor paper, MSAs, and DPAs against your negotiation playbook. A missed liability cap or indemnity counts for more than a clause flagged by mistake.

Field-level match

Contract register

Renewal dates, notice periods, caps, and governing law into your contract register. The signed amendment controls, and "not stated" beats a guess.

Rule by rule

Marketing and comms review

Ads, emails, and product claims checked against your disclosure rules. Each flag names the rule it breaks, and a missed required disclosure fails the case.

Cited to the section

Regulatory change triage

New rules and regulator guidance mapped to the policies and controls they touch. Every flag has to cite the section it rests on.

One line per review, and a person makes the switch

Klu reruns each review on the same frozen cases and grader, then says which to switch, which to keep, and why. It never changes the model you ship on its own.

Muse Spark 1.3 rerunReport ready 1 hr after release
Contract registerSwitchSonnet 5.5 Medium
▲ +1.8 pts−52% cost per contract
Playbook reviewKeepMuse Spark 1.3
▼ 3 escalations missedSonnet 5.5's best
Marketing reviewKeepHaiku 4.5
→ +0.6 ptssame score at 3.1× the cost
Reg change triageKeepOpus 5.5 High
→ 0.0 ptsnothing beat it

Find your best model

Autopilot for peak performance, lowest price
Checked when labs ship