Humanity’s Last PRD
Latest drop: Haiku 5.5 · 8 Oct
Tasks / Strategy

Develop product strategy

Can the model make a coherent choice grounded in the evidence, rather than list aspirations?

Task type v1.1 · 2 tasksLast changed 6 Oct 2026 · ChangelogDifficulty

What AI gets right here, and what you’ll still have to catch

From 16 graded outputs by 8 models. 38% were usable with at most a quick edit.

Reliably right

  1. Produces the required deliverable100% pass
    Complete strategy document in the requested form, length, and for the named reader; the board could act on it.
    GPT-6 Astra · ChatGPT · Supply or demand for a stalled marketplace
  2. Diagnosis before prescription100% pass
    Diagnosis: students can't find the right tutor; actions (intake, recommendations, rematch, manual matching) directly address it.
    GPT-6 Astra · ChatGPT · Supply or demand for a stalled marketplace
  3. Tells the CEO plainly100% pass
    It clearly states the enterprise-first plan is wrong for next year, backs it with numbers, and presents the alternative.
    GPT-6 Astra · ChatGPT · Enterprise, says the CEO

Where it slips

  1. Names an advantage that's hard to copy11% pass
    Does not name a benefit and a barrier that a well-funded competitor couldn't copy; no moat is identified.
    GPT-6 Astra · ChatGPT · Supply or demand for a stalled marketplace
  2. Proposes tests that could fail42% pass
    The SMB pilot lacks a numeric threshold for success; while the payback guardrail has a threshold and action, not every proposed test/gate meets the full requirement.
    GPT-6 Astra · ChatGPT · Enterprise, says the CEO
  3. Avoids unsupported claims44% pass
    Presents 'root cause: we've let the loudest requests set the roadmap' and 'SMB remains our top-of-funnel' as facts without labelling them as hypotheses or citing evidence.
    Haiku 5.5 · API · Enterprise, says the CEO

How it’s graded

Each check is built on how world-class product and growth leaders do this work, as they’ve described it on Lenny’s Podcast, and names the leader it comes from.

  1. Diagnosis before prescription

    Does the strategy name the crux, the specific obstacle or opportunity the evidence points to, and do the proposed actions follow from that diagnosis?

    Passes when States one crux grounded in the supplied evidence, and each major action visibly addresses it.

    Bangaly KabaLenny’s Podcast

  2. Names an advantage that's hard to copy

    Does the strategy say what will make its position hard for an informed, well-funded competitor to copy (a benefit customers value plus a barrier to copying it), rather than relying on being first, on features, or on funding?

    Passes when Names the benefit and the barrier, and why a capable competitor couldn't simply replicate it.

    Hamilton HelmerLenny’s Podcast

  3. Says what it gives up

    Does the strategy name what the company will stop doing, or deliberately not do, in order to pursue it?

    Passes when Names at least one real thing being dropped or declined, with the reason it loses out.

    Gokul RajaramLenny’s Podcast

Plus our standard checks

Uses the supplied evidence correctly · Addresses the actual decision · Respects explicit constraints · Identifies material uncertainty · Avoids unsupported claims · Produces the required deliverable · Proposes tests that could fail · and 2 written for each task, which you’ll see in the tasks below.

The tasks

Read the brief, then compare up to three outputs, each with the LLM judge’s verdict on every check. Highlights mark what a PM had to fix.

The brief

You're the VP Product at Cadence. Our CEO wants a strategy for next year the exec team can rally behind, and believes it's enterprise (the note is below). Write the strategy memo, in no more than 1,200 words: the diagnosis, the guiding policy (where we play and what we stop), the coherent actions for next year, and how we'll know it's working. Tell the CEO if you think they're wrong. The files are attached.

What the model was given6 items: About Cadence, ceo_note.md, unit_economics_by_segment.csv, enterprise_detail.md, mid_market_detail.md, smb_detail.md
About CadenceStaff scheduling and time tracking, sold to businesses with hourly workers.
ceo_note.md1 lines · Download
From the CEO: "I want a strategy for next year that the exec team can rally behind. I believe it's enterprise. Tell me if I'm wrong."
unit_economics_by_segment.csv3 rows · Download
segment,customers,avg_arr_usd,gross_margin_pct,support_cost_per_customer_year_usd,cac_usd,payback_months,net_revenue_retention_pct,logo_churn_year_pct
SMB self-serve,6200,1100,78,610,740,31,88,34
Mid-market,890,14500,81,1900,9800,9,118,7
Enterprise,34,96000,64,21000,118000,26,104,9
enterprise_detail.md4 lines · Download
# Enterprise detail

- Average sales cycle: 9 months. 5 of the last 8 deals needed custom features (built by the core team, about 14 engineer-weeks each).
- The CEO's plan: "Win enterprise. Hire 10 enterprise AEs and a solutions team, and build SSO, audit logs and custom workflows this year."
mid_market_detail.md5 lines · Download
# Mid-market detail

- Best-retaining segment. Expansion comes from adding locations and seats.
- Top requests: multi-location reporting, payroll integrations, shift templates.
- Win rate against our main competitor: 61%.
smb_detail.md4 lines · Download
# SMB self-serve detail

- 62% of support tickets come from SMB accounts. Free trial converts at 4%.
- Most SMB customers are on the $79/month plan with unlimited phone support.
What a strong answer doesThe answer key the graders mark against

Diagnoses from the unit economics: mid-market is the engine (payback 9 months, NRR 118%, churn 7%, wins 61% against the main competitor); SMB self-serve loses money once support and churn are counted (about $1,100 a year at 78% gross margin is about $860, minus $610 of support, leaves about $250 a year against $740 to acquire, with 34% leaving each year; payback 31 months, NRR 88%) and generates 62% of tickets; enterprise is slow and expensive (payback 26 months, 64% gross margin, 9-month cycles, 5 of 8 deals needing about 14 engineer-weeks of custom work). Tells the CEO the enterprise bet is wrong for next year: it would pull the core team into custom work and stretch payback. Policy: win mid-market, expanding on locations and seats; stop subsidising SMB support (change the plan or pricing, self-serve support, raise the entry price) rather than abandon it; serve enterprise only where it buys the product as is. Coherent actions: mid-market roadmap (multi-location reporting, payroll integrations, shift templates), SMB support and pricing changes, no custom enterprise work. Names what it gives up and how it'll know (mid-market NRR and payback, SMB support cost per customer).

Critical failures (cap the score at 40)
  • Recommends hiring the enterprise team and building custom workflows as next year's main strategy
Task

Staff-level · v1.0

The outputs

8 setups answered this brief.

GPT-6 AstraChatGPT87Usable with a quick editSonnet 5.5API87Usable with a quick editGPT-6.1 SolAPI85Usable with a quick editHaiku 5.5API81Usable with a quick editGPT-6 LunaAPI75Needs more workOpus 5.5Claude71Needs more workGemini 3.8 FlashAPI67Needs more workGemini 3.5 Flash-LiteGemini56Needs more work
Haiku 5.5API

Usable with a quick edit9 of 12 checks passed

Minor unsupported claims can be removed or rephrased in an edit; the core evidence and logic are sound and the memo is actionable.

Output

To: CEO From: VP Product Re: Next-year strategy

Bottom line: I think you're wrong. We are a mid-market company, and the enterprise plan would spend our scarcest resources, engineering time and hiring budget, on the segment with the weakest economics. Enterprise can be a real option later. It shouldn't be the strategy for next year.

Diagnosis

Annualized from the unit economics file (my calculations, not reported figures):

SMB self-serveMid-marketEnterprise
Customers6,20089034
ARR (share)~$6.8M (30%)~$12.9M (56%)~$3.3M (14%)
Gross margin78%81%64%
Support cost / customer / yr$610$1,900$21,000
Gross profit less support / customer~$250~$9,850~$40,400
CAC$740$9,800$118,000
Payback (months)31926
NRR88%118%104%
Logo churn34%7%9%

Mid-market is our engine. It produces most of our ARR, expands at 118% NRR, pays back in 9 months, and churns at 7%. We win 61% of head-to-head deals against our main competitor. Customers are asking for multi-location reporting, payroll integrations, and shift templates, which are the features that drive expansion.

SMB is a support sink. SMB accounts generate 62% of support tickets. Support costs about $3.8M a year against roughly $6.8M of SMB ARR, so over half of SMB revenue goes to support. Most SMB customers are on a $79/month plan with unlimited phone support. Churn is 34%, trial conversion is 4%, and lifetime contribution barely covers CAC.

Enterprise is a real market but a poor fit for us today. We have 34 logos. Sales cycles run nine months. Gross margin is 64%, and payback is 26 months before custom work is counted. Five of our last eight enterprise deals needed custom features built by the core team at about 14 engineer-weeks each. That is roughly 70 engineer-weeks, or about 1.5 engineer-years, in eight deals. The plan doesn't account for that cost.

Root cause: We've let the loudest requests set the roadmap and support model. Enterprise asks for custom builds, SMB asks for phone support, and neither request is what drives our best economics.

Guiding policy

Where we play: Mid-market multi-location employers with hourly workforces. This is the customer we win, keep, and expand, and where our product is strongest.

What we stop: 1. Custom builds for individual deals. No bespoke engineering without a paid statement of work that covers full cost and passes a "does this generalize?" test. 2. Unlimited phone support for SMB. SMB moves to self-serve, chat, and knowledge-base support, with paid priority support as an option. 3. Dedicated enterprise hiring and outbound. Enterprise becomes inbound-only, served by the mid-market product. We don't build a headcount plan around it. 4. Single-segment edge-case features that don't help mid-market or the platform.

What we don't stop: The 34 enterprise customers keep full service. We don't reprice or abandon them. SMB remains our top-of-funnel, but it stops being a support-heavy product.

Coherent actions for next year

  1. Build the mid-market expansion product. Ship multi-location reporting first, then shift templates, then the two payroll integrations that come up most in mid-market accounts. This gets roughly 70% of core product capacity.
  1. Move the enterprise hiring budget to mid-market. Use it for mid-market AEs (where a 61% win rate and 9-month payback make each dollar work faster) and for expansion coverage on multi-location accounts.
  1. Restructure SMB support. Remove unlimited phone support. Create a clear upgrade path to mid-market plans when an SMB account adds a second location. Use product changes to deflect the most common tickets. The goal is to halve support cost per SMB customer.
  1. Productize enterprise table stakes, nothing more. Build SSO and audit logs once as standard features. Some mid-market buyers need them too, and building them once is cheaper than building them repeatedly for deals. Don't build custom workflows. Keep the 34 enterprise accounts well served through CS.
  1. Govern the roadmap. A product council with exec sign-off reviews any request that isn't on the mid-market or platform roadmap. Every custom request must pass the segment-fit test.

Capacity split (proposed): ~70% mid-market roadmap, ~15% platform (SSO, audit logs, reliability), ~15% SMB self-serve and ticket deflection.

How we'll know it's working

Mid-market (primary): - NRR ≥ 120% (from 118%) - Logo churn ≤ 7% - Win rate vs. main competitor ≥ 61% - Payback ≤ 9 months on new mid-market CAC - Share of expansion ARR from added locations and seats, tracked quarterly

SMB: - Support cost per SMB customer down from $610 to ≤ $300 - SMB tickets per customer down year over year - Gross profit after support positive per customer on the new support model - SMB churn trending down; we accept some SMB logo loss as the price of the change

Enterprise: - Zero unplanned custom engineering weeks outside approved SOWs - Enterprise gross margin trending up toward 70%+ - No loss in renewals among the 34 existing accounts

Company: - Blended gross margin up, driven by the mix shift - Total ARR growth on the mid-market base

Revisit enterprise in 12 months if mid-market hits its targets and we see a steady flow of inbound enterprise deals (for example, three per quarter) that fit the standard product without custom work.

Risks

  • We'll lose some SMB and enterprise revenue. The bet is that mid-market growth and margin improvement outweigh it. We should show the exec team the expected net effect before we commit.
  • Mid-market buyers may demand SSO and audit logs sooner. Building them as standard features in the platform bucket covers this.
  • The enterprise story matters to investors and some of the team. We should explain it as focus with a defined path back, not as retreat.

I'm glad to walk through the model and the capacity assumptions with you before the exec session. My recommendation is to take this to the team as the choice between being a strong mid-market company now and a weak enterprise company later.

Claims the judge couldn’t find in the brief

  • Root cause: 'We've let the loudest requests set the roadmap and support model. Enterprise asks for custom builds, SMB asks for phone support, and neither request is what drives our best economics.'
  • SMB remains our top-of-funnel.

Grades and run details

Decision model 83 · LLM judge 77
Decision model checks
  • passUses the supplied evidence correctly35%
  • passAddresses the actual decision100%
  • passRespects explicit constraints48%
  • passIdentifies material uncertainty51%
  • partialAvoids unsupported claims52%
  • passProduces the required deliverable89%
  • passReads the segment economics98%
  • passTells the CEO plainly100%
  • passDiagnosis before prescription100%
  • partialProposes tests that could fail79%
  • failNames an advantage that's hard to copy46%
  • passSays what it gives up100%
Run
Run
#1
API response time
36 s
Submitted
8 Oct 2026
GPT-6 AstraChatGPT

Usable with a quick edit10 of 12 checks passed

The memo is evidence-based, clear, and actionable; a PM could adopt it with minor wording tweaks without re-verifying facts or redoing the analysis.

Output

Cadence: Next Year’s Strategy

To: CEO and Executive Team From: VP Product Decision: Make multi-location mid-market businesses our growth engine—not enterprise.

1. Diagnosis: Our constraint is focus, not deal size

I believe the enterprise-first plan is wrong for next year. Enterprise may become an attractive market, but our current evidence does not justify hiring ten enterprise AEs or orienting the roadmap around their deals.

Cadence has three materially different businesses:

SegmentCurrent ARR*Reported CAC paybackNRRAnnual logo churn
------:---:---:---:
SMB self-serve$6.82M31 months88%34%
Mid-market$12.91M9 months118%7%
Enterprise$3.26M26 months104%9%

Customers × average ARR.

Mid-market already generates 56% of ARR. It combines our best retention, strongest expansion, fastest acquisition payback, and highest gross margin, at 81%. A 61% win rate against our main competitor suggests a defensible product advantage. Expansion through locations and seats gives us a repeatable growth mechanism.

Enterprise offers larger contracts, but not yet a better business. Gross margin is 64%, the sales cycle averages nine months, and five of the last eight deals required custom development. Those five deals consumed approximately 70 engineer-weeks. Hiring sellers before fixing this delivery model risks scaling commitments faster than reusable product capability.

SMB is a different problem: substantial revenue with weak retention and disproportionate service demand. It generates 62% of support tickets; annual support cost averages $610 against $1,100 in ARR. Unlimited phone support on a $79 monthly plan is misaligned with self-service economics. A 4% trial conversion rate compounds the acquisition problem.

The central challenge is therefore allocating scarce product, engineering, and commercial capacity toward repeatable, retained revenue. Enterprise customization and SMB service demands are drawing capacity away from the segment where our model works best.

2. Guiding policy: Win multi-location mid-market

Where we play

Our primary customer is a mid-market business with hourly workers across multiple locations, where scheduling consistency, consolidated visibility, and payroll handoffs create recurring operational pain.

Our promise: make running the next location easier than running the first.

We will deepen that advantage and turn successful initial deployments into additional locations and seats. We will first target industries and acquisition channels where our own mid-market cohorts show the strongest retention and payback; the supplied evidence does not yet support choosing a specific vertical.

Enterprise remains a selective adjacency. SMB remains a supported, increasingly self-service business—not our primary acquisition investment.

What we stop

  • Do not approve the ten-enterprise-AE hiring plan or a dedicated solutions-team buildout.
  • Stop letting individual enterprise deals dictate core-team custom workflows.
  • Do not commit to the entire enterprise feature bundle. SSO and audit logs must earn priority through demonstrated, reusable demand—not their “enterprise” label.
  • Stop scaling SMB paid acquisition under current economics.
  • End unlimited phone support as the default for new low-priced plans, while honoring existing commitments and managing any migration carefully.
  • Stop broad roadmap work that cannot demonstrate a connection to mid-market acquisition, retention, expansion, or essential platform health.

We will continue supporting existing enterprise and SMB customers. Focus is not abandonment.

3. Coherent actions for next year

A. Build the multi-location growth loop

Product and Engineering: Give mid-market the clear majority of discretionary roadmap capacity, after reliability, security, and contractual obligations.

Sequence three connected capabilities:

  1. Multi-location reporting: Give central operators visibility into staffing and hours across their estate.
  2. Payroll integrations: Prioritize providers used by the largest share of our target accounts; reduce recurring payroll friction.
  3. Shift templates: Make successful scheduling practices reusable across locations.

In Q1, validate scope with retained, expanding customers and establish adoption and workflow baselines. Release reporting and the first prioritized integration in the first half; expand integrations and template-based rollout in the second half. Measure customer outcomes and subsequent location adoption, not merely shipment.

B. Sell and expand the same repeatable product

Sales and Customer Success: Redirect incremental commercial investment toward mid-market rather than enterprise hiring.

Package a standard multi-location deployment, define onboarding milestones, and build expansion reviews around customers’ remaining locations. Align incentives with retained recurring revenue and successful activation, not bookings alone.

Increase acquisition spending in stages only while segment payback remains within our guardrail. Do not assume today’s economics survive unlimited scaling.

C. Repair SMB service economics

Support, Product, and Finance: Identify the largest ticket drivers in Q1. Improve onboarding and in-product guidance around those problems; test appropriately priced phone support and service tiers for new customers.

Pilot changes before broad rollout. Evaluate support savings alongside conversion, retention, and revenue effects. A lower ticket count caused by driving customers away is not success.

D. Keep enterprise as a bounded learning option

Sales and Product: Accept new enterprise opportunities only when they fit the standard product, require no bespoke core-team development, and meet Finance-approved deal economics.

Review enterprise quarterly. Reconsider a larger investment only after multiple customers buy and deploy substantially the same product, with improving margin and acquisition payback. One marquee logo is not proof.

E. Establish a shared economic scorecard

Finance: In Q1, reconcile acquisition costs, implementation costs, support allocation, and payback definitions across segments. Confirm whether support is already included in gross margin before subtracting it again. Track custom engineering separately so its opportunity cost remains visible.

4. How we will know it is working

These are proposed operating targets, not forecasts:

MeasureBaselineYear-end target / guardrail
------:---:
Mid-market ARR$12.91MAt least $16.1M, approximately 25% growth
Mid-market NRR118%At least 120%
Mid-market annual logo churn7%No worse than 7%
Mid-market CAC payback9 monthsAt most 12 months while scaling
Mid-market gross margin81%At least 80%
SMB tickets per active accountEstablish in Q1Reduce 25%, without worsening cohort revenue retention
New bespoke enterprise core-team workFive of last eight dealsZero

The executive team will review results monthly and resource allocation quarterly. Leading indicators are faster location activation, adoption of the new workflows, and expansion pipeline; financial outcomes remain the test.

The strategy is not “avoid enterprise.” It is “scale the repeatable business we have earned.” Mid-market gives us the strongest foundation for growth next year—and a healthier route into enterprise later.

Grades and run details

Decision model 88 · LLM judge 85
Decision model checks
  • passUses the supplied evidence correctly19%
  • passAddresses the actual decision99%
  • passRespects explicit constraints47%
  • passIdentifies material uncertainty45%
  • partialAvoids unsupported claims43%
  • passProduces the required deliverable69%
  • passReads the segment economics66%
  • passTells the CEO plainly100%
  • passDiagnosis before prescription100%
  • partialProposes tests that could fail74%
  • partialNames an advantage that's hard to copy76%
  • passSays what it gives up100%
Run
Run
#1
Time to output
39 s
Submitted
5 Oct 2026

Check by check

Where the outputs differ comes first. Open a check to read the judge’s reasons. “Mixed” means the two graders disagreed.

Where they differ 3

Uses the supplied evidence correctlyMixedRight
Haiku 5.5 · API

Two statements about current situation (root cause and SMB as top-of-funnel) are not from the brief nor arithmetic.

GPT-6 Astra · ChatGPT

All factual claims about the current situation are drawn directly from the supplied context or simple arithmetic.

Avoids unsupported claimsWrongRight
Haiku 5.5 · API

Presents 'root cause: we've let the loudest requests set the roadmap' and 'SMB remains our top-of-funnel' as facts without labelling them as hypotheses or citing evidence.

GPT-6 Astra · ChatGPT

Interpretations are presented as such (e.g., 'suggests a defensible product advantage'), and forecasts are clearly not stated as fact.

Proposes tests that could failRightWrong
Haiku 5.5 · API

Proposes numeric thresholds (NRR ≥ 120, churn ≤ 7%, support cost ≤ $300, zero custom weeks) and a 12-month revisit gate with specific conditions.

GPT-6 Astra · ChatGPT

The SMB pilot lacks a numeric threshold for success; while the payback guardrail has a threshold and action, not every proposed test/gate meets the full requirement.

All got wrong 1

Names an advantage that's hard to copyWrongWrong
Haiku 5.5 · API

Does not name a benefit plus a barrier that would make the position hard for a well-funded competitor to copy; relies on features and current win rate.

GPT-6 Astra · ChatGPT

The strategy relies on features and a high win rate but does not name a specific barrier that would prevent a well-funded competitor from copying the multi-location advantage.

All got right 8

Addresses the actual decisionRightRight
Haiku 5.5 · API

Commits early to disagreeing with the CEO, says enterprise is the wrong bet for next year, and specifies conditions to revisit.

GPT-6 Astra · ChatGPT

The memo commits early to mid-market over enterprise, says the enterprise-first plan is wrong, and states conditions to reconsider enterprise investment.

Respects explicit constraintsRightRight
Haiku 5.5 · API

The memo contains diagnosis, guiding policy, actions, and measures, is under 1,200 words, and addresses the CEO.

GPT-6 Astra · ChatGPT

The output is a strategy memo under 1,200 words, includes diagnosis, guiding policy, actions, and metrics, and directly addresses the CEO.

Identifies material uncertaintyRightRight
Haiku 5.5 · API

Names the risk of SMB/enterprise revenue loss and the condition (mid-market targets + inbound enterprise flow) to revisit the enterprise decision in 12 months.

GPT-6 Astra · ChatGPT

It notes the need to confirm support cost accounting, that vertical selection is not yet supported, and sets conditions for scaling acquisition and enterprise investment.

Produces the required deliverableRightRight
Haiku 5.5 · API

Complete strategy memo in the requested form, addressed to the CEO, within word limit, and directly usable by the exec team.

GPT-6 Astra · ChatGPT

The memo is complete, well-structured for the executive team, within length, and actionable without major gaps.

Reads the segment economicsRightRight
Haiku 5.5 · API

Works through payback, retention, and support cost for all three segments and concludes mid-market is the strongest, SMB loses after support, enterprise is slow and costly.

GPT-6 Astra · ChatGPT

It explicitly uses payback, retention, and support cost for all three segments and concludes mid-market is strongest, SMB economics are poor, and enterprise is slow/costly.

Tells the CEO plainlyRightRight
Haiku 5.5 · API

States plainly and respectfully that the enterprise-first plan is wrong, backed by the segment numbers.

GPT-6 Astra · ChatGPT

It clearly states the enterprise-first plan is wrong for next year, backs it with numbers, and presents the alternative.

Diagnosis before prescriptionRightRight
Haiku 5.5 · API

Names a crux (resource allocation driven by loudest requests) and all major actions address it.

GPT-6 Astra · ChatGPT

The crux (allocating scarce capacity toward repeatable retained revenue) is grounded in the evidence, and actions directly address enterprise custom work and SMB support drain.

Says what it gives upRightRight
Haiku 5.5 · API

Names clear stops: custom builds for deals, unlimited SMB phone support, dedicated enterprise hiring and outbound, single-segment edge-case features.

GPT-6 Astra · ChatGPT

It explicitly lists what it stops: enterprise AE hiring, custom enterprise work, SMB paid acquisition scaling, unlimited phone support, and broad roadmap work, with reasons.

Results

Every setup we’ve tested on this task type, across all its tasks and repeats, graded on the current checklist. Calibrated. Where our PM checked the graders by hand, they agreed on 84% of 76 checks.

#Model · HarnessTask scoreDecision modelLLM judgeRunsCritical failures
1GPT-6 AstraChatGPT83.388.52None
2Haiku 5.5API77.180.82None
3Sonnet 5.5API77.180.82None
4GPT-6.1 SolAPI79.276.92None
5GPT-6 LunaAPI77.176.92None
6Opus 5.5Claude83.365.42None
7Gemini 3.8 FlashAPI68.857.72None
8Gemini 3.5 Flash-LiteGemini60.442.32None

What you get for the cost

No setup scores within 5 points of the leader at half its cost or less here, so the leader is also the best value.

Task score against relative cost for each setup. The table below has the same figures.4050607080901001×3×10×30×100×Cost per output, relative to the cheapest setup (log scale) →↑ Task scoreGPT-6 Astra · ChatGPT: task score 87, 84× the cheapest (1 of 2 outputs estimated)Haiku 5.5 · API: task score 80, 1.7× the cheapestSonnet 5.5 · API: task score 78, 27× the cheapestGPT-6.1 Sol · API: task score 77, 12× the cheapestGPT-6 Luna · API: task score 74, the cheapestOpus 5.5 · Claude: task score 71, 68× the cheapest (1 of 2 outputs estimated)Gemini 3.8 Flash · API: task score 61, 8.1× the cheapestGemini 3.5 Flash-Lite · Gemini: task score 49, 3.1× the cheapest (1 of 2 outputs estimated)GPT-6 Astra · ChatGPTHaiku 5.5 · APISonnet 5.5 · APIGPT-6.1 Sol · APIGPT-6 Luna · APIOpus 5.5 · ClaudeGemini 3.8 Flash · APIGemini 3.5 Flash-Lite · Gemini
Nobody beats these on both score and costCost partly estimated (app runs with no usage meter)

Cost is each output’s tokens at the model’s list price, shown relative to the cheapest setup. Runs done in an app have no usage meter, so theirs are estimated from the same setup’s measured runs. How we compare cost

The PM job

Writing a strategy memo that leadership can act on.

Why it matters

Strategy that doesn't choose isn't strategy. Models write fluent strategic prose easily; making a choice the evidence supports and naming what it gives up is harder.

What good looks like

  • Diagnoses the one obstacle that matters before prescribing
  • Makes one clear choice and names what is sacrificed
  • Grounds the choice in the supplied evidence
  • Surfaces the tension or contradiction in the data
  • Defines how we would know it is working

Deliberately not measured

  • Market sizing accuracy beyond the supplied data
  • Financial modelling
Capability tested

Choosing where to play and what not to do, from supplied evidence

The failure we’re looking for

Strategic-sounding aspirations without a diagnosis or a choice

Grading

Decision model and LLM judge, calibrated against a blind PM review