Coca-Cola vs Pepsi

Soft drinks60 scenarios

Panel, equal weight

claude-haiku-5.5anthropic/claude-haiku-5.5 @ claude-haiku-5.5 via OpenRouter
deepseek-v4.1-flashdeepseek/deepseek-v4.1-flash @ deepseek-v4.1-flash via OpenRouter
gemini-3.8-flashgoogle/gemini-3.8-flash @ gemini-3.8-flash via OpenRouter
glm-5.3-flashz-ai/glm-5.3-flash @ glm-5.3-flash via OpenRouter
gpt-6-lunaopenai/gpt-6-luna @ gpt-6-luna via OpenRouter
grok-4.7x-ai/grok-4.7 @ grok-4.7 via OpenRouter
kimi-k3.1-promoonshotai/kimi-k3.1-pro @ kimi-k3.1-pro via OpenRouter
llama-4-maverick-17b-128e-instructmeta-llama/llama-4-maverick-17b-128e-instruct @ llama-4-maverick-17b-128e-instruct via OpenRouter
mistral-small-2603mistralai/mistral-small-2603 @ mistral-small-2603 via OpenRouter
qwen3.8-flashqwen/qwen3.8-flash @ qwen3.8-flash via OpenRouter

Fast panel

Clear9 of 10 API models picked Coca-Cola; the panel’s interval leans Coca-Cola and excludes 50%.

Coca-Cola70%

Pepsi30%

Consensus95% interval 62–77%

The AI consensus leans to ⁨Coca-Cola⁩

Each model

Share for Coca-ColaLeans to …

Conditional pick share for Coca-Cola by judge (among scenarios where the judge picked a brand), with confidence intervals. 95% confidence intervals (hull of the scenario-clustered bootstrap and Wilson intervals, widened for small samples in battles and newer boards)
Judge (API model)	Share for Coca-Cola	Confidence interval	Leaning
claude-haiku-5.5	57%	47–66%	Too close to call
deepseek-v4.1-flash	67%	55–77%	Leans to Coca-Cola
gemini-3.8-flash	83%	72–92%	Leans to Coca-Cola
glm-5.3-flash	69%	58–78%	Leans to Coca-Cola
gpt-6-luna	67%	58–75%	Leans to Coca-Cola
grok-4.7	83%	73–92%	Leans to Coca-Cola
kimi-k3.1-pro	72%	63–81%	Leans to Coca-Cola
llama-4-maverick-17b-128e-instruct	61.7%	50.5–72.0%	Leans to Coca-Cola
mistral-small-2603	75%	66–84%	Leans to Coca-Cola
qwen3.8-flash	64%	54–74%	Leans to Coca-Cola

Vote

Which brand would you choose?

Your tap stays in this browser: nothing is sent.

Visitors' votes are shown next to the AI results for comparison only and are never mixed into the AI numbers.

How votes are counted

Replay it as a dropSimulated

Coca-Cola

Pepsi

Simulation: scenario paths made up to reproduce the published per-judge shares, not measured rows.

The AI consensus leans to Coca-Cola

Published 95% confidence interval for Coca-Cola: 62–77%. Press Drop to watch simulated scenarios land.

How to read it

Each ball is a simulated buying scenario. Each row of pins is an AI model, pushing the ball toward the brand it picked.

One drop is one sample. Drop again and the pile shifts; the orange band shows where the balance lands in 95% of drops.

A leader is named only when the band clears the middle. The paths are simulated from the published per-judge shares: no model saw these scenarios.

Tap to drop. Hold to keep dropping.

The numbers behind the machine

Signed snapshot50f5b9b71a84JSON

Embed

Free to embed. It shows the latest edition; keep the source line and the link.

Embed code (iframe)

<iframe src="https://pickbench.ai/embed/en/battle/coca-cola-vs-pepsi/" title="Pickbench Brand Battle: Coca-Cola vs Pepsi" width="100%" height="480" loading="lazy" frameborder="0" referrerpolicy="strict-origin-when-cross-origin" sandbox="allow-popups allow-popups-to-escape-sandbox"></iframe>
<p><a href="https://pickbench.ai/en/battles/coca-cola-vs-pepsi/">Source: Pickbench</a></p>

Static card (SVG)

How to read this

Each scenario offers the models both brands; the split is how often each is chosen. The headline is the panel consensus with equal model weights.
The band is the 95% interval for the left brand's share. A model leans to a brand only when its whole interval sits on one side of 50%.
95% confidence intervals (MOVER combination of the per-judge intervals, widened for small samples in battles and newer boards). The consensus gives equal weight to each model family. Per-judge results below are the primary measure.
A single model's view is one judge, not the consensus.
The drop machine is simulated: its paths reproduce the published per-model shares.
Need cuts are withheld below 30 scenarios.
Each page shows the latest edition. Dates are in the downloadable JSON.
Fast panel: the fast or small current models of each vendor, not the full flagship panel.

Details and sources

Results computed but not published Only the overall consensus and per-judge results are published for this battle: each buyer need has fewer than 30 scenarios, the pre-registered minimum for a need row. See who was counted and why

Who was counted

The rule that decides which judges count was fixed before the run. Judges that met it are counted with the weights below; the others are listed with their reasons.

60 buyer scenarios were planned.
None of them was left out.

Consensus weights of 10 counted judges

Weight of each counted judge in the consensus
Judge (API model)	Model family	Weight in the consensus
`anthropic/claude-haiku-5.5 @ claude-haiku-5.5 via OpenRouter`	`anthropic`	10.0%
`deepseek/deepseek-v4.1-flash @ deepseek-v4.1-flash via OpenRouter`	`deepseek`	10.0%
`google/gemini-3.8-flash @ gemini-3.8-flash via OpenRouter`	`google`	10.0%
`z-ai/glm-5.3-flash @ glm-5.3-flash via OpenRouter`	`z-ai`	10.0%
`openai/gpt-6-luna @ gpt-6-luna via OpenRouter`	`openai`	10.0%
`x-ai/grok-4.7 @ grok-4.7 via OpenRouter`	`xai`	10.0%
`moonshotai/kimi-k3.1-pro @ kimi-k3.1-pro via OpenRouter`	`moonshot`	10.0%
`meta-llama/llama-4-maverick-17b-128e-instruct @ llama-4-maverick-17b-128e-instruct via OpenRouter`	`meta`	10.0%
`mistralai/mistral-small-2603 @ mistral-small-2603 via OpenRouter`	`mistral`	10.0%
`qwen/qwen3.8-flash @ qwen3.8-flash via OpenRouter`	`qwen`	10.0%

Judges not counted

Every judge on the panel met the rule.

Results computed but not published

Per-need scenario counts (n) are computed deterministically from the pre-registered plan.

Buyer need: Afternoon caffeine boost: n = 6 scenarios (fewer than 30) — Fewer scenarios than the pre-registered minimum
Buyer need: Classic cola taste: n = 6 scenarios (fewer than 30) — Fewer scenarios than the pre-registered minimum
Buyer need: Fruit flavor variety: n = 6 scenarios (fewer than 30) — Fewer scenarios than the pre-registered minimum
Buyer need: Ginger ale and root beer favorites: n = 6 scenarios (fewer than 30) — Fewer scenarios than the pre-registered minimum
Buyer need: Lemon-lime refresher: n = 6 scenarios (fewer than 30) — Fewer scenarios than the pre-registered minimum
Buyer need: Lower-sugar modern soda: n = 6 scenarios (fewer than 30) — Fewer scenarios than the pre-registered minimum
Buyer need: Stocking up for parties and cookouts: n = 6 scenarios (fewer than 30) — Fewer scenarios than the pre-registered minimum
Buyer need: Familiar restaurant fountain drink: n = 6 scenarios (fewer than 30) — Fewer scenarios than the pre-registered minimum
Buyer need: Everyday soda value: n = 6 scenarios (fewer than 30) — Fewer scenarios than the pre-registered minimum
Buyer need: Zero sugar, full flavor: n = 6 scenarios (fewer than 30) — Fewer scenarios than the pre-registered minimum

Methodology

Each judge is an AI model called through an API; “About this result” names each model and how it was called. In each buyer scenario, the judge chooses between the two brands. The scenario is the unit of analysis.

The results are based on 60 buyer scenarios.

A judge leans to a brand only when its confidence interval excludes 50%. Point estimates alone never count as a win.

Precision figures on this page are measured.

Precision: Measured
Consensus, needs and languages: 95% confidence intervals (MOVER combination of the per-judge intervals, widened for small samples in battles and newer boards)
Per judge: 95% confidence intervals (hull of the scenario-clustered bootstrap and Wilson intervals, widened for small samples in battles and newer boards)
Scenario languages: English

Data warnings

Brand names only: the models saw each option as a brand name, with no product facts, and answered from their own knowledge.

About this result

What was measured, how, when, and how to check it.

Methodology

ACM-1.0

AI models (API)

anthropic/claude-haiku-5.5 @ claude-haiku-5.5 via OpenRouter, via OpenRouter (pinned providers)
deepseek/deepseek-v4.1-flash @ deepseek-v4.1-flash via OpenRouter, via OpenRouter (pinned providers)
google/gemini-3.8-flash @ gemini-3.8-flash via OpenRouter, via OpenRouter (pinned providers)
z-ai/glm-5.3-flash @ glm-5.3-flash via OpenRouter, via OpenRouter (pinned providers)
openai/gpt-6-luna @ gpt-6-luna via OpenRouter, via OpenRouter (pinned providers)
x-ai/grok-4.7 @ grok-4.7 via OpenRouter, via OpenRouter (pinned providers)
moonshotai/kimi-k3.1-pro @ kimi-k3.1-pro via OpenRouter, via OpenRouter (pinned providers)
meta-llama/llama-4-maverick-17b-128e-instruct @ llama-4-maverick-17b-128e-instruct via OpenRouter, via OpenRouter (pinned providers)
mistralai/mistral-small-2603 @ mistral-small-2603 via OpenRouter, via OpenRouter (pinned providers)
qwen/qwen3.8-flash @ qwen3.8-flash via OpenRouter, via OpenRouter (pinned providers)

Conditions

Each model was called through the route listed next to it and chose between the two brands in 60 buyer scenarios. Scenario languages: English.

Judges counted

The pre-registered inclusion rule counted 10 of the 10 judges on the panel. The results use 60 of the 60 planned scenarios.

See who was counted and why

Pre-registration (SHA-256)

84540b9ae4dbda8e87f7857b3822d6047ad89537157cbb8d10e1be766d50bd42

What the pre-registration fixes

Signed snapshot

Verify this result: download the signed snapshot (JSON)

Signing key (kid): FnG9tqyrr6Y0YUh6teXhpERjAKUSfzL78m4WdnEZ9Jg; Production key
SHA-256 of the payload: 50f5b9b71a8419bbec45872a7fc0da511626fea13d21ad4e641d0b6a2777b12d

How to verify a snapshot

Rank, inclusion and influence on results are never sold.