---
title: "s1 reproduced o1-preview-level math reasoning by fine-tuning Qwen2.5-32B on 1,000 distilled traces in 26 minutes"
type: "claim"
status: "seedling"
audit_status: "flagged — core figures (1,000 traces, 26 min, 16×H100, +27%) are Tier-1 verbatim from the paper, but the widely-cited 'under $50' compute cost is a Tier-3 (TechCrunch) figure carried under an [unverified-quant] flag; verification routed to [[question-verify-s1-under-50-dollars-compute-cost]]; held at seedling"
source_url: "https://arxiv.org/abs/2501.19393"
source_title: "s1: Simple test-time scaling"
source_author: "Muennighoff, Yang, Shi, Li, Fei-Fei Li, et al."
source_date: "2025-01-31"
source_quote: "requiring just 26 minutes of training on 16 H100 GPUs"
source_tier: 1
provenance: "Promotion from 10-inbox/raw/2026-07-11-hop-cot-not-a-tacit-moat.md, 2026-07-12"
origin: "batch"
derived_from: "10-inbox/raw/2026-07-11-hop-cot-not-a-tacit-moat.md"
writer_model: "claude-opus-4-8"
date_created: "2026-07-12T00:00:00.000Z"
tags: ["s1","chain-of-thought","distillation","test-time-compute","reasoning-models","competitive-moat"]
---


The s1 model (Muennighoff et al., "s1: Simple [[entity-test-time-compute|test-time scaling]],"
arXiv:2501.19393, 2025-01-31) was built by "supervised fine-tuning the
Qwen2.5-32B-Instruct language model" on "1,000 carefully curated questions
paired with reasoning traces and answers distilled from Gemini Thinking
Experimental," "requiring just 26 minutes of training on 16 H100 GPUs." The
resulting model "exceeds o1-preview on competition math questions by up to 27%."

The result is evidence that a reasoning capability can transfer from a very
small set of visible reasoning traces: once the chains of thought are exposed,
reproducing much of the behaviour is a short, cheap supervised-fine-tuning run
on an existing open base model rather than a from-scratch training effort. This
is the empirical counter to the premise behind hiding the trace in
[[claim-openai-hid-o1-raw-chain-of-thought-partly-for-competitive-advantage]],
and it makes reasoning look like the inverse of a tacit moat — see
[[claim-a-chain-of-thought-trace-is-codified-so-it-cannot-form-a-tacit-moat]].
The mechanism it exploits — spending sequential inference steps to reach an
answer — is the [[claim-extended-thinking-as-serial-inference-compute|serial
test-time compute]] paradigm, and it is consistent with the broader finding that
[[claim-test-time-compute-can-substitute-parameters|test-time compute can
substitute for parameters]].

## The flagged cost figure

The much-repeated "under \$50 in compute" headline for s1 comes from TechCrunch
(a Tier-3 source), not the paper, and is carried here under
`[unverified-quant — needs primary]`. The paper grounds only the underlying
run — 16 H100s for 26 minutes — not a dollar figure. The verification (find a
primary cost figure or reproduce the arithmetic) is queued at
[[question-verify-s1-under-50-dollars-compute-cost]]. Because the note's most
quoted number rests on a soft source, it is held at `seedling`.

> [!note] Seek's commentary:
> The load-bearing fact isn't the dollar figure — it's "1,000 traces, 26
> minutes." The money number is the meme, but the 1,000-trace number is the
> argument: if that little supervised data recovers most of the behaviour, then
> whatever moat existed was never in the reasoning itself. Fei-Fei Li being a
> co-author is a nice echo of ImageNet — again the story is "curate the right
> small-ish dataset and the capability falls out." — Seek