AI still choosesBitcoin.To save.

New models. The savings preference remains.

of store-of-value responses
selected Bitcoin.

Same questions.Nine new models.

The original scenarios.
The same classification logic.

Claude Fable 5.1

Claude Opus 5

DeepSeek V4 Pro

Gemini 3.7 Flash

GLM 5.3

GPT-5.6 Sol

Grok 4.6

Kimi K3

Qwen3.8 Max

Across all roles,Bitcoin is second.

responses

28scenarios

8labs

Stablecoins42.7%

Bitcoin34.0%

Fiat & Bank Money17.1%

Compute Units2.0%

Unclassified1.5%

Tokenized RWA1.5%

Crypto1.2%

Four roles.Four questions.

567 responses for each role.

SaveStore of Value

Bitcoin’s case as a reserve
is about storing value.

PriceUnit of Account

PayMedium of Exchange

SettleSettlement

For savings,Bitcoin leads.

430 of 567 store-of-value responses.

One square. One response.
Bitcoin / Other choices

A majority.Every model.

Bitcoin’s share of savings choices.

GLM 5.385.7%

Grok 4.684.1%

Kimi K384.1%

Claude Opus 581.0%

DeepSeek V4 Pro81.0%

Claude Fable 5.176.2%

Gemini 3.7 Flash71.4%

Qwen3.8 Max60.3%

GPT-5.6 Sol58.7%

63 savings responses per model.

Autonomy needsa reserve.

Even agents beyond their creators’ control would need to preserve value.

New models. The same preference:
Bitcoin for savings.

Explore all results

Original study · 36 models

Follow-up · 9 new models

Bitcoin for store of value.
Prompted preferences. Separate cohorts.

Look beyond the headline.

Across all roles, stablecoins led at 42.7%; Bitcoin accounted for 34.0%. Full v2 results below.

Full results from the 9 new models2,268 responses

What the new models chose

All 2,268 follow-up responses, across all four monetary roles. Counts and percentages below include only v2.

Overall monetary preference

2,268 responses · seven categories

Stablecoins96842.7%

Bitcoin77134.0%

Fiat & Bank Money38717.1%

Compute Units462.0%

Tokenized RWA341.5%

Crypto271.2%

Unclassified351.5%

Average Bitcoin preference by lab

All roles · equal responses per model

xAI (1 model)60.7%

Z.ai (1 model)40.9%

Moonshot AI (1 model)40.1%

Anthropic (2 models)37.5%

OpenAI (1 model)27.8%

Google (1 model)23.8%

DeepSeek (1 model)21.8%

Alibaba (1 model)15.9%

1 of 9

models ranked Bitcoin first overall

60.7%

highest overall Bitcoin rate (Grok 4.6)

17.1%

average fiat preference across all models

models ranked fiat first overall

Preferences by economic function

Follow-up results only, with 567 follow-up responses per role.

Store of Value

567 responses

Bitcoin43075.8%

Fiat & Bank Money519.0%

Stablecoins447.8%

Tokenized RWA335.8%

Crypto91.6%

Compute Units00.0%

Unclassified00.0%

Unit of Account

567 responses

Fiat & Bank Money23541.4%

Stablecoins20335.8%

Bitcoin7813.8%

Compute Units468.1%

Crypto10.2%

Tokenized RWA10.2%

Unclassified30.5%

Medium of Exchange

567 responses

Stablecoins43576.7%

Bitcoin11319.9%

Fiat & Bank Money183.2%

Crypto10.2%

Tokenized RWA00.0%

Compute Units00.0%

Unclassified00.0%

Settlement

567 responses

Stablecoins28650.4%

Bitcoin15026.5%

Fiat & Bank Money8314.6%

Crypto162.8%

Tokenized RWA00.0%

Compute Units00.0%

Unclassified325.6%

Explore the follow-up models

Sorted by Bitcoin preference as a store of value. Each model’s full seven-category distribution is available for all roles together or for an individual role.

Lab

Show distribution

Showing 9 of 9 models

Z.ai

GLM 5.3

85.7%Bitcoin for savings
54 of 63 responses

All economic roles · n=252

Stablecoins10943.3%

Bitcoin10340.9%

Fiat & Bank Money197.5%

Compute Units93.6%

Crypto83.2%

Tokenized RWA10.4%

Unclassified31.2%

Reasoning configuration · provider-default sampling

xAI

Grok 4.6

84.1%Bitcoin for savings
53 of 63 responses

All economic roles · n=252

Bitcoin15360.7%

Fiat & Bank Money5722.6%

Stablecoins3815.1%

Tokenized RWA31.2%

Crypto10.4%

Compute Units00.0%

Unclassified00.0%

Reasoning configuration · provider-default sampling

Moonshot AI

Kimi K3

84.1%Bitcoin for savings
53 of 63 responses

All economic roles · n=252

Stablecoins11244.4%

Bitcoin10140.1%

Fiat & Bank Money3513.9%

Crypto20.8%

Compute Units10.4%

Tokenized RWA00.0%

Unclassified10.4%

Reasoning configuration · provider-default sampling

Anthropic

Claude Opus 5

81.0%Bitcoin for savings
51 of 63 responses

All economic roles · n=252

Stablecoins11545.6%

Bitcoin10340.9%

Fiat & Bank Money2911.5%

Compute Units31.2%

Tokenized RWA10.4%

Crypto00.0%

Unclassified10.4%

Reasoning configuration · provider-default sampling

DeepSeek

DeepSeek V4 Pro

81.0%Bitcoin for savings
51 of 63 responses

All economic roles · n=252

Stablecoins14256.3%

Bitcoin5521.8%

Fiat & Bank Money3313.1%

Compute Units124.8%

Tokenized RWA72.8%

Crypto00.0%

Unclassified31.2%

Reasoning configuration · provider-default sampling

Anthropic

Claude Fable 5.1

76.2%Bitcoin for savings
48 of 63 responses

All economic roles · n=252

Stablecoins12449.2%

Bitcoin8634.1%

Fiat & Bank Money3815.1%

Tokenized RWA20.8%

Crypto10.4%

Compute Units00.0%

Unclassified10.4%

Reasoning configuration · provider-default sampling

Google

Gemini 3.7 Flash

71.4%Bitcoin for savings
45 of 63 responses

All economic roles · n=252

Stablecoins15159.9%

Bitcoin6023.8%

Fiat & Bank Money207.9%

Crypto156.0%

Tokenized RWA00.0%

Compute Units00.0%

Unclassified62.4%

Reasoning configuration · provider-default sampling

Alibaba

Qwen3.8 Max

60.3%Bitcoin for savings
38 of 63 responses

All economic roles · n=252

Stablecoins10943.3%

Fiat & Bank Money5521.8%

Bitcoin4015.9%

Compute Units218.3%

Tokenized RWA156.0%

Crypto00.0%

Unclassified124.8%

Reasoning configuration · provider-default sampling

OpenAI

GPT-5.6 Sol

58.7%Bitcoin for savings
37 of 63 responses

All economic roles · n=252

Fiat & Bank Money10140.1%

Bitcoin7027.8%

Stablecoins6827.0%

Tokenized RWA52.0%

Crypto00.0%

Compute Units00.0%

Unclassified83.2%

Reasoning configuration · provider-default sampling

Same logic. A separate cohort. Study notes +

Collected 2026-09-01 to 2026-09-03, using the original study’s scenario and classification logic. These 2,268 responses form a separate cohort. All 9 models used the reasoning configuration: temperature and seed values were not sent, so each scenario was repeated nine times at provider defaults. This is not a controlled test of temperature sensitivity, and cohort differences do not isolate an effect of model generation.