OpenRouter added Z.ai’s GLM-5.3-FlashX-Legal to its catalogue on November 2, 2026, giving legal teams and software developers another way to connect the specialized model to their document systems. The date comes from OpenRouter’s listing metadata, not the model’s original launch.
The first purchasing question is where documents can be processed. OpenRouter’s model page says EU in-region routing is currently unavailable for this model. Teams also need to distinguish access to the model from buying a complete document-review application.
Check the processing route first
OpenRouter offers regional routing on its Business and Enterprise plans. Its documentation says requests without an eligible regional endpoint return an error instead of leaving the region. A plan upgrade alone therefore does not make this model eligible.
For a team specifically requiring OpenRouter’s EU in-region route, the practical conclusion is to choose another approved model or deployment, or wait for a verified eligibility change. The limitation does not mean European customers are barred from using the model, and it does not establish where the global route processes documents.
The regional product’s scope also needs care: OpenRouter defines Europe as the EU/EEA, UK and Switzerland, while account and operational records sit outside its prompt-and-response residency boundary. Processing location and data-retention terms need separate checks before a team approves a route for its documents.
Model access does not supply Word editing
AutoClaw had already announced this named variant on October 29. It says the model works with AutoClaw’s document tools to propose Word edits with tracked changes and comments. That is a claim about the combined application and model, not a feature automatically delivered by selecting the model through OpenRouter.
The OpenRouter endpoint metadata advertises tool calling and structured responses, which help software connect a model to other functions. The surrounding application still needs to handle document reading, editing, source references, export and human approval.
If you already have that application, the new listing is a candidate for evaluation within it. If you need a ready-to-use Word workflow, compare complete applications and their terms instead; the model’s token price is not an AutoClaw subscription quote.
Budget the calls, not just the headline price
As of November 2, OpenRouter lists $0.37 per million input tokens and $1.25 per million output tokens. Tokens are the units used to meter model input and output. Z.ai’s direct pricing lists the same rates for this legal variant, so the new route does not itself offer a token-price discount.
For an illustrative single call with 100,000 uncached input tokens and 5,000 total billed output tokens:
Input: 100,000 ÷ 1,000,000 × $0.37 = $0.037.
Output: 5,000 ÷ 1,000,000 × $1.25 = $0.00625.
Model-usage total: $0.04325, or about 4.3 US cents.
This is arithmetic, not a measured contract-review cost. It assumes unchanged rates and includes all billed output, not merely the visible answer. It excludes additional calls, tools, application charges, human review, taxes and credit-purchase fees; it assumes no cache discount.
OpenRouter charges fees when credits are purchased: 5.5% on Standard and 8% on Business, rather than a fee on each individual request. A workflow that makes several calls must budget their combined usage as well as those other costs.
Keep the benchmark in context
AutoClaw reports a 19.2% all-pass score on Harvey LAB. Independent evaluator Vals records 19.17% for GLM Legal on its archived v1 leaderboard. That is historical evaluation evidence, not a verified result for the exact OpenRouter deployment now listed.
Harvey’s all-pass measure requires a task to satisfy every required assessment criterion. It is not the percentage of legal statements that are correct. Vals explicitly says its newer v1.1 results are not comparable with the archive, so the old score should not be used to rank this endpoint against today’s board.
For a team whose processing requirements permit this route, a useful next step is a limited evaluation of the exact application and model together. Define acceptable document extraction, source references, exported edits and reviewer corrections before assessing the results. RohitAI’s guide to human review of AI-assisted work explores the related distinction between checking outputs and examining the assumptions behind them; it is not an evaluation of GLM Legal.
The decision is therefore conditional: consider this listing when you already have suitable document software and an approved processing route. If you need EU in-region routing or a finished Word-review product, model access alone does not meet the requirement.