惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
Jina AI
Jina AI
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
T
Troy Hunt's Blog
T
The Exploit Database - CXSecurity.com
Microsoft Security Blog
Microsoft Security Blog
V
Visual Studio Blog
F
Fortinet All Blogs
博客园_首页
P
Proofpoint News Feed
V
Vulnerabilities – Threatpost
The Cloudflare Blog
C
Cyber Attacks, Cyber Crime and Cyber Security
cs.CL updates on arXiv.org
cs.CL updates on arXiv.org
H
Heimdal Security Blog
cs.AI updates on arXiv.org
cs.AI updates on arXiv.org
A
About on SuperTechFans
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
AI
AI
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
S
Security Affairs
The Register - Security
The Register - Security
S
Security @ Cisco Blogs
Hugging Face - Blog
Hugging Face - Blog
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
博客园 - 聂微东
Schneier on Security
Schneier on Security
WordPress大学
WordPress大学
Google DeepMind News
Google DeepMind News
GbyAI
GbyAI
T
Tailwind CSS Blog
Hacker News: Ask HN
Hacker News: Ask HN
W
WeLiveSecurity
D
Docker
L
LangChain Blog
B
Blog RSS Feed
The Last Watchdog
The Last Watchdog
Cloudbric
Cloudbric
TaoSecurity Blog
TaoSecurity Blog
N
Netflix TechBlog - Medium
酷 壳 – CoolShell
酷 壳 – CoolShell
I
InfoQ
The Hacker News
The Hacker News
AWS News Blog
AWS News Blog
Exploit-DB.com RSS Feed
Exploit-DB.com RSS Feed
宝玉的分享
宝玉的分享
I
Intezer
云风的 BLOG
云风的 BLOG
V2EX - 技术
V2EX - 技术
K
KPMG report finds enterprise disconnect between AI and its ROI | CIO

Simon Willison's Weblog

Release: datasette 1.0a29 Thoughts on GitLab’s workforce reduction A quote from James Shore Your AI Use Is Breaking My Brain TIL: Using LLM in the shebang line of a script Learning on the Shop floor A quote from New York Times Editors’ Note A quote from Andrew Quinn A quote from Luke Curley Release: llm-gemini 0.31 Tool: Big Words Behind the Scenes Hardening Firefox with Claude Mythos Preview Notes on the xAI/Anthropic data center deal Tool: GitHub Repo Stats Live blog: Code w/ Claude 2026 Vibe coding and agentic engineering are getting closer than I’d like Release: datasette-referrer-policy 0.1 Release: datasette-llm 0.1a7 Release: llm-echo 0.5a0 Granite 4.1 3B SVG Pelican Gallery A quote from Andy Masley April 2026 newsletter Research: TRE Python binding — ReDoS robustness demo Tool: Redis Array Playground A quote from Anthropic Sightings iNaturalist Sightings Codex CLI 0.128.0 adds /goal Our evaluation of OpenAI's GPT-5.5 cyber capabilities Quoting Andrew Kelley We need RSS for sharing abundant vibe-coded apps Release: llm 0.32a1 LLM 0.32a0 is a major backwards-compatible refactor Release: llm 0.32a0 Quoting OpenAI Codex base_instructions Quoting Matthew Yglesias What's new in pip 26.1 - lockfiles and dependency cooldowns! Introducing talkie: a 13B vintage language model from 1930 microsoft/VibeVoice Tracking the history of the now-deceased OpenAI Microsoft AGI clause WHY ARE YOU LIKE THIS Quoting Romain Huet GPT-5.5 prompting guide llm 0.31 DeepSeek V4 - almost on the frontier, a fraction of the price Tool: Millisecond Converter It's a big one russellromney/honker Serving the For You feed Extract PDF text in your browser with LiteParse for the web A pelican for GPT-5.5 via the semi-official Codex backdoor API Release: llm-openai-via-codex 0.1a0 Quoting Maggie Appleton A quote from Bobby Holley Is Claude Code going to cost $100/month? Probably not—it’s all very confusing Where’s the raccoon with the ham radio? (ChatGPT Images 2.0) A quote from Andreas Påhlsson-Notini scosman/pelicans_riding_bicycles Release: llm-openrouter 0.6 TIL: SQL functions in Google Sheets to fetch data from Datasette Claude Token Counter, now with model comparisons Headless everything for personal AI Research: Claude system prompts as a git timeline Adding a new content type to my blog-to-newsletter tool - Agentic Engineering Patterns Join us at PyCon US 2026 in Long Beach—we have new AI and security tracks this year Release: datasette 1.0a28 Release: llm-anthropic 0.25 Tool: datasette.io news preview Release: datasette-export-database 0.3a1 Release: datasette 1.0a27 Gemini 3.1 Flash TTS Tool: Gemini 3.1 Flash TTS A quote from Kyle Kingsbury Release: datasette-ports 0.3 Zig 0.16.0 release notes: “Juicy Main” datasette PR #2689: Replace token-based CSRF with Sec-Fetch-Site header protection Tool: SQLite Query Result Formatter Demo Tool: SQLite Query Result Formatter Demo A quote from Giles Turnbull A quote from Giles Turnbull Research: SQLite WAL Mode Across Docker Containers Sharing a Volume Research: SQLite WAL Mode Across Docker Containers Sharing a Volume Tool: Cleanup Claude Code Paste Release: datasette-ports 0.1 Eight years of wanting, three months of building with AI A quote from Chengpeng Mou Tool: Syntaqlite Playground Release: scan-for-secrets 0.2 Release: scan-for-secrets 0.1.1 Release: scan-for-secrets 0.1 Release: research-llm-apis 2026-04-04 A quote from Kyle Daigle Vulnerability Research Is Cooked The cognitive impact of coding agents A quote from Willy Tarreau A quote from Daniel Stenberg A quote from Greg Kroah-Hartman Research: Can JavaScript Escape a CSP Meta Tag Inside an Iframe? The Axios supply chain attack used individually targeted social engineering Highlights from my conversation about agentic engineering on Lenny’s Podcast
Qwen3.6-35B-A3B on my laptop drew me a better pelican than Claude Opus 4.7
Simon Willison · 2026-04-17 · via Simon Willison's Weblog

16th April 2026

For anyone who has been (inadvisably) taking my pelican riding a bicycle benchmark seriously as a robust way to test models, here are pelicans from this morning’s two big model releases—Qwen3.6-35B-A3B from Alibaba and Claude Opus 4.7 from Anthropic.

Here’s the Qwen 3.6 pelican, generated using this 20.9GB Qwen3.6-35B-A3B-UD-Q4_K_S.gguf quantized model by Unsloth, running on my MacBook Pro M5 via LM Studio (and the llm-lmstudio plugin)—transcript here:

The bicycle frame is the correct shape. There are clouds in the sky. The pelican has a dorky looking pouch. A caption on the ground reads Pelican on a Bicycle!

And here’s one I got from Anthropic’s brand new Claude Opus 4.7 (transcript):

The bicycle frame is entirely the wrong shape. No clouds, a yellow sun. The pelican is looking behind itself, and has a less pronounced pouch than I would like.

I’m giving this one to Qwen 3.6. Opus managed to mess up the bicycle frame!

I tried Opus a second time passing thinking_level: max. It didn’t do much better (transcript):

The bicycle frame is entirely the wrong shape but in a different way. Lines are more bold. Pelican looks a bit more like a pelican.

I don’t think Qwen are cheating

A lot of people are convinced that the labs train for my stupid benchmark. I don’t think they do, but honestly this result did give me a little glint of suspicion. So I’m burning one of my secret backup tests—here’s what I got from Qwen3.6-35B-A3B and Opus 4.7 for “Generate an SVG of a flamingo riding a unicycle”:

I’m giving this one to Qwen too, partly for the excellent <!-- Sunglasses on flamingo! --> SVG comment.

What can we learn from this?

The pelican benchmark has always been meant as a joke—it’s mainly a statement on how obtuse and absurd the task of comparing these models is.

The weird thing about that joke is that, for the most part, there has been a direct correlation between the quality of the pelicans produced and the general usefulness of the models. Those first pelicans from October 2024 were junk. The more recent entries have generally been much, much better—to the point that Gemini 3.1 Pro produces illustrations you could actually use somewhere, provided you had a pressing need to illustrate a pelican riding a bicycle.

Today, even that loose connection to utility has been broken. I have enormous respect for Qwen, but I very much doubt that a 21GB quantized version of their latest model is more powerful or useful than Anthropic’s latest proprietary release.

If the thing you need is an SVG illustration of a pelican riding a bicycle though, right now Qwen3.6-35B-A3B running on a laptop is a better bet than Opus 4.7!