😺 OpenAI shipped GPT-5.5 today

发布时间：2026-04-23来源：The Neuron

Your browser does not support the audio element.

Welcome, humans.

Add "timber" to the list of jobs AI is coming for.

Weyerhaeuser,

America's largest landowner

, is using AI to digitize its entire forest inventory and pursue autonomous logging equipment, aiming to double profits by 2030 without needing lumber prices to lift a finger.

If a tree falls in the woods and only a neural net is there to hear it, does it make a sound?

Anyway, the rest of the AI world spent the week trying to figure out why Opus 4.7 suddenly feels like a GPT model and GPT-5.5 feels like Claude.

Uno reverse day. Opposite universe. Parallel dimension.

Let's get into it.

Here’s what happened in AI today:

😼 OpenAI shipped GPT-5.5 one week after Anthropic's Opus 4.7

📰 Meta cut 8,000 jobs (10% of staff) to fund its AI push

📰 Anthropic quietly hit a $1 trillion valuation, passing OpenAI

🍪 Spotify now lives inside Claude for personalized music recs

💡 This week's best AI reads, from Mollick to Dylan Patel to Andreessen

… and a

whole lot more that you can read about here

.

P.S:

Want to reach 675,000 AI-hungry readers?

Click here to advertise with us.

P.P.S:

Love robots? We’re starting a new robotics newsletter!

Sign up early here

.

😼

OpenAI shipped GPT-5.5 to put Anthropic on notice.

OpenAI dropped

GPT-5.5

yesterday, exactly seven days after Anthropic shipped

Opus 4.7

. We're officially in a one-lab-one-week launch cadence, and nobody is blinking.

Here's what happened:

GPT-5.5 is "worker-class," meaning built to finish tasks instead of just answering questions. It's live in ChatGPT and Codex for Plus, Pro, Business, and Enterprise; API access comes "very soon" at $5/$30 per million input/output tokens.

It wins Terminal-Bench 2.0 (82.7% vs Opus 4.7's 69.4%) and ties or beats industry pros on 84.9% of

GDPval

tasks across 44 jobs. It loses SWE-Bench Pro (58.6% vs 64.3%) with an asterisk citing Anthropic's own flagged "

signs of memorization

" on that eval.

It jumps from 27.1% to 35.4% on FrontierMath Tier 4 and helped discover a

new proof

about off-diagonal Ramsey numbers, later verified in Lean.

OpenAI rated it "High" on both bio/chem and cyber capability. Partner

XBOW

called it "Mythos-like hacking, open to all," prompting

Trusted Access for Cyber

for vetted defenders.

Why this matters:

This is the first clean "GPT beats Claude" moment in over a year, and it landed exactly seven days after Anthropic's best. If you run agents in production, you probably want to re-test them on both models this weekend.

Our take:

Full disclosure: Corey's been a longtime GPT fan and stayed with them; Grant has not.

Testing 5.5 was the first time in forever (since o3, maybe) Grant actually liked using a GPT model. He never got on board with the 5 series. He'd begrudgingly use 5.1 and 5.2 only when Claude rate-limited him. For the last 18 months he's been a power Claude user.

5.5 hasn't pushed him all the way over,

but there's something much nicer about using this one

. It doesn't write too much (unlike most "smart" GPTs before it). It doesn't sound as dumb when thinking fast. And honestly, it feels a little bit like Claude.

Meanwhile Opus 4.7 feels like the opposite: it feels like a GPT. More tokens, clunkier to talk to, Claude's signature vibe harder to find.

Are we in uno reverse land? Is it opposite day? Did we enter a parallel universe?

There's actually a real explanation. Every's Dan Shipper

noticed the same thing

: Opus 4.7 "feels slow" next to 5.5 because OpenAI has a hardware advantage you can actually feel. And SemiAnalysis' Dylan Patel

pointed out

that Anthropic silently went from an L4 engineer (Opus 4.6) to an L6 (Mythos) in two months. The Opus 4.7 the rest of us get is compute-starved and deliberately restrained.

Anthropic is a Ferrari on fuel rationing; OpenAI just bought the gas station.

(No offense meant.)

FROM OUR PARTNERS

AI is moving from copilots to autonomous systems, and enterprises need infrastructure built for that shift.

The

Dell AI Factory with NVIDIA

delivers a validated, end-to-end AI stack spanning infrastructure, software, and services, designed to help organizations operationalize agentic AI at scale.

Built for production:

Up to 269% ROI in year one.

Up to 86% faster AI deployment.

Flexible deployment across on-prem, edge, and hybrid environments.

Learn More

🎓

AI Skill of the Day: The Opus 4.7 + GPT-5.5 handoff.

Every's Dan Shipper just found

the cheat code

for senior-engineer-level coding: use Opus 4.7 to write the plan, then hand that plan to GPT-5.5 to execute. On Every's internal Senior Engineer benchmark, this combo scored 62.5/100 (for context: human senior engineers score 80-90; Opus 4.7 alone scores in the low 30s; GPT-5.5 alone scores in the low-to-mid 40s).

Why it works: Opus 4.7 writes tight, contract-style plans (exact file counts, line limits, first-principles design). That turseness is off-putting in chat, but it's exactly what gives GPT-5.5 the confidence to delete files, rewrite from scratch, and carry a multi-hour refactor through end-to-end instead of patching around the mess.

To try it: open Claude with Opus 4.7 selected and ask it to write a rewrite plan for your target codebase. Then paste that plan into Codex or ChatGPT with GPT-5.5 selected, and say this:

Here is a plan written by a senior engineer for rewriting this codebase from first principles. Execute it faithfully. Do not patch around the existing code: delete what the plan says to delete, rewrite what it says to rewrite, and match its conceptual structure exactly. Carry the plan through from start to finish.

One favorite insight: if you only have 5.5, add explicit "here's what good looks like" targets (e.g., "this file should be ~100 lines when done") to draw out

its boldness

.

Want more tips like this? Check out our

AI Skill of the Day Digest for April

.

Total AI beginner?

Start here

(

goes with this video

).

Have a specific skill you want to learn?

Request it here.

🍪 Treats to Try

*Asterisk = from our partners (only the first one!).

Advertise to 675K+ readers here

!

Claude's new everyday connectors

hook AllTrails, Instacart, Audible,

, and TripAdvisor into your chat; find a hike, order groceries, book a trip, or pick an audiobook without leaving the window. Free with any Claude plan.

Reloop Animation Studio

turns a script into a full AI-generated video ad in Ultra-Realistic, Pixar, Manga, or 3D Clay style with character and scene consistency end-to-end, no prompts needed; no pricing details.

AirJelly

watches what you're working on and flags overdue tasks and unresolved intents before you ask; pitched as the first proactive desktop agent of its kind; no pricing details.

Google Stitch

open-sourced the

DESIGN.md

spec (plus a wizard to extract one from your product) so any coding agent can import or export your visual identity system instead of guessing; free to try.

Noscroll

is an AI bot that does your doomscrolling for you and texts you a high-signal daily digest over SMS, so you can kill the feed apps; no pricing details.

📰 Around the Horn

Meta

cut 8,000 jobs (10% of staff) to fund its AI push (

Bloomberg

).

Anthropic

quietly hit a $1 trillion valuation on secondary markets, passing OpenAI.

The White House

accused China of "industrial-scale" AI theft via

distillation

(copying a competitor's model by training on its outputs).

The Pentagon

vibe-coded 100,000+ AI agents on Gemini for unclassified government work.

Google

unveiled eighth-generation TPUs, two new chips built for the agentic era.

Freshfields

is helping Amazon and Anthropic build legal AI tools that Anthropic will then resell to rival law firms.

Want absolutely EVERYTHING that happened in AI this week? Click here!

FROM OUR PARTNERS

AI Beyond the Pilot Phase

AI has everyone talking. Not everyone has answers. At Gladly Connect Live, CX leaders from Condé Nast, Smith Optics, and more share exactly how they moved AI from pilot to production, the timeline, the systems, the QA loops. 13+ sessions built for the moment we're all in. For CX and ecommerce leaders. Atlanta, May 4–6. Space is limited, secure your spot now.

💡

Intelligent Insights:

Sign of the Future

(Ethan Mollick, One Useful Thing): why GPT-5.5 is less a single model and more a data point on the capability curve you should pay attention to.

GPT-5.5 Vibe Check

(Dan Shipper, Every): a three-week hands-on review that lands on a counterintuitive workflow: Opus 4.7 as planner, GPT-5.5 as executor.

The Supply and Demand of AI Tokens

(Dylan Patel, SemiAnalysis, with Patrick O'Shaughnessy): why Anthropic's revenue is exploding while its compute quietly isn't, and what that means for who gets access to the frontier.

Monitoring the Situation

(Marc Andreessen, a16z): a unified theory of how the internet reinvented the 2.5-day outrage cycle, and why the past was never as peaceful as we remember it.

OpenAI's Cofounders Finally Tell Their Side

(Sam Altman and Greg Brockman with Ashlee Vance and Kylie Robison, Core Memory): first joint podcast, covering the Elon trial, Sora getting cut, and Sam's worst week.

Software Fundamentals Matter More Than Ever

(Matt Pocock, AI Hero): a conference talk on why "specs-to-code" is vibe-coding in disguise, and what actually makes AI-written codebases last.