Claude Opus 5 vs Qwen 3.7 Max
for reading a P&L statement

The bottom line
Claude explains it while Qwen counts

Neither model can see why a number moved, because a P&L only shows the outcome. Claude Opus 5 comes out ahead when the job is naming the real cause behind a margin swing, and Qwen 3.7 Max is the cheaper option for turning the statement into a clean variance table first.

That split comes from the closest evidence available rather than from a benchmark built for this exact job. Claude Opus 5 scores 58.6 percent on Finance Agent, a test built from real financial-research questions⁸, and reaches 1,735 Elo on GDPval-AA v2, an evaluation of finished professional work that includes finance, against 1,190 Elo for Qwen 3.7 Max⁹. Both scores are about broader financial judgment, not margin-bridge analysis specifically, so read them as a strong lean and not a settled verdict.

A basis point is one hundredth of a percentage point, and it is how finance teams usually describe a margin move, so a shift from 12.0 percent operating margin to 10.5 percent is a 150 basis-point drop. In a staged workflow, Qwen 3.7 Max can parse the statement, calculate the dollar, percentage and basis-point changes and rank the candidates. Claude Opus 5 is the stronger reviewer of whether those candidates really explain the margin move, and the one to write the final explanation a decision-maker reads.

Who this is for
Which finance roles this fits

What we compared
The models not the accounting app

This page compares Claude Opus 5 and Qwen 3.7 Max through their API in one neutral setup, not one model wrapped in one accounting product against the other in a different one.

The task is a pasted or uploaded P&L with no connected accounting platform behind it, so the two periods, the subtotals and the margins all come from the text or file you give the model. What matters is whether the model reconciles the numbers, calculates the changes correctly, ranks the line items that explain most of the margin move, and says plainly when the statement alone cannot prove the cause.

We left tool rows out of the specs table on purpose. A file-upload button, a spreadsheet add-on or a connected general ledger depends on the app wrapped around the model, so the same model can look stronger or weaker depending on the product it sits in. Judging that here would compare wrappers, not the two models.

Specs at a glance
The numbers that matter for a P&L

Both models have far more context and output room than an ordinary P&L needs. The numbers below are the ones that change how a margin analysis gets built and priced.

Head to head
Where each model earns the edge

The dimension-by-dimension read, mapped to the report's own comparisons. Cost and speed favor Qwen 3.7 Max. Explaining the driver behind the move favors Claude Opus 5.

How to test
A fair test on your own statements

A useful test looks boring on purpose. Same statement, same prompt, same reasoning setting, same calculator, same output schema for both models. Then judge whether the answer reconciled, ranked the real drivers and needed little correction.

What the evidence shows
Good proxies for a different task

No public benchmark grades a model on reading a pasted P&L and naming the margin driver. Here is what the closest available evidence measures.

How to prompt each one
The same task needs different prompts

Claude and Qwen reward different prompt shapes for the same P&L, so matching the prompt to the model does more for quality than the model choice by itself.

Give Claude Opus 5 the full P&L first, then a compact instruction that sets the scope, the evidence boundary and a word limit. Anthropic says Opus 5 already self-verifies, so repeated double-check instructions can push it into over-verifying instead of helping¹⁴.

Give Qwen 3.7 Max an explicit calculation sequence and a required output schema, and turn on thinking mode for the analysis pass. Alibaba recommends thinking for detailed report analysis, and the explicit steps help stop a response that just paraphrases the biggest-looking variance⁷.

A Claude Opus 5 prompt: scope and a word limit

Analyse the P&L below. Lead with the two line items that
explain most of the operating-margin change. Build a
basis-point bridge from prior-period margin to current-period
margin. Separate observed drivers from possible operational
causes. Do not infer price, volume or mix unless the statement
supports it. Keep the executive explanation under 180 words.

A Qwen 3.7 Max prompt: an explicit calculation sequence

Parse the P&L into rows. For each row calculate current value,
prior value, dollar change, percentage change, current percent
of revenue, prior percent of revenue and basis-point contribution
to the operating-margin change. Reconcile the bridge. Then rank
the top three drivers and explain them. Label every statement
as OBSERVED or HYPOTHESIS.

Weak spots
Too much prose or too little structure

Neither model is perfect on this task. The useful question is where each one adds cleanup work, and what to change in the prompt or the workflow.

Which one to choose
Start with who reads the answer

One question first: is the output going straight to a decision-maker, or is it a first pass across many statements? Follow the branch that matches most of your work.

Recommendations
Where each profile should start

If the output goes straight to a decision-maker who needs the real cause of a margin swing, start with Claude Opus 5, and require explicit observed-versus-hypothesis labels when the statement is ambiguous.

If this is first-pass processing across many statements, or cost and interactive speed matter most, start with Qwen 3.7 Max and route only the unusual, high-value or low-confidence cases to Claude.

When the P&L arrives with budgets, cost-centre schedules and commentary attached, start with Claude Opus 5 since synthesis matters more, even though both models have enough context for the extra material. When the output must be strict JSON for a reporting pipeline, either model can work, so run a schema-compliance test and choose Qwen if quality comes out equal. For anything that affects an earnings release, an investment decision or a board paper, use Claude to draft and still require human accounting review before it goes out.

Bottom line
The safer explanation still costs more

Claude Opus 5 is the better single-model choice for explaining what moved the margin. Qwen 3.7 Max is the better cost-performance choice for calculating and structuring the evidence before that explanation gets written.

Both models can calculate a line change well enough. Claude's advantage tends to show up further along, when the largest variance is not the real driver, when several movements offset each other, or when the honest answer is that the P&L does not reveal the underlying operational cause.

The limits here are real. There is no public exact-model benchmark for pasted P&L margin attribution, finance benchmarks use different data and harnesses, and Qwen's strongest finance result comes from a specialised architecture built around it rather than the base model alone. Both models have also moved on, since Alibaba shipped Qwen 3.8 Max as its new flagship and Anthropic positions Fable 5.1 above Opus 5¹³.

The safest final step is to test the shape of your own statements, not a generic prompt from the internet. A fair test needs the same setup for both models, the same P&L, the same prompt and the same place to run them, so the result reflects the models and not the tool around them. In practice that is harder than it sounds, since most teams end up running one model in one app and the other in a different one, on two separate subscriptions, which tilts the comparison before the first answer comes back. The cleaner the setup, the more the difference you see is really Claude Opus 5 vs Qwen 3.7 Max, and not just which one happened to be easier to reach that day.

Frequently asked
questions

Related comparisons