Development

Four-billion-parameter models match last generation's eighty

development2026-03-03

Claim

Qwen released four open models matching prior 80-billion-parameter performance at just 4 billion, all runnable on phones, while Cognition's SWE-1.6 reached near-Opus coding performance at 950 tokens per second on a hundredfold more RL compute.

Reported metric: 4B ≈ 80B

Reporting date: 2026-03-03 (not necessarily the event date)

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

Welcome to March 3, 2026 (theinnermostloop.substack.com)

Knowledge graphopen in browser →

Reported in

Welcome to March 3, 2026

Actor

Evidence for

Supersedes

The smallest adder falls from 121 parameters to 36 in a week

Referenced by

Cite or correct this record

Citation formatCitation text

Four-billion-parameter models match last generation's eighty
Innermost Loop KB — secondary record
Reporting date: 2026-03-03
KB permalink: https://nicholsn.github.io/innermost-loop-kb/developments/2026-03-03-qwen-4b-matches-80b
Claim: Qwen released four open models matching prior 80-billion-parameter performance at just 4 billion, all runnable on phones, while Cognition's SWE-1.6 reached near-Opus coding performance at 950 tokens per second on a hundredfold more RL compute.
Attributed sources:
- Welcome to March 3, 2026 — https://theinnermostloop.substack.com/p/welcome-to-march-3-2026
This citation does not imply independent verification of linked-source claims.

Suggest a correction — opens a GitHub draft for you to review and submit.