Skip to main content
Back

OpenAI GPT-5.5: 82.7% Terminal-Bench, $5/M tokens, live now

OpenAI ships GPT-5.5, Qwen 27B beats 397B on coding, MIT model handles 10M tokens, Google open-sources AI design guidelines

April 24, 2026 6 min read

Hey,

OpenAI had a big day. GPT-5.5 drops with stronger agentic coding, and Workspace Agents ship to automate team workflows in Slack. Two launches, one clear message: the prompt era is over.

We're moving from "ask AI" to "AI just handles it." MIT's 10M-token recursive model quietly reinforces the same shift from a research angle. Context is no longer the bottleneck.

OpenAI is building a full operating system for work. That's the real story.
Together with · partner_image
Author photo
Today's Author
Lior Alexander. Founder of AlphaSignal and former ML Engineer. Previously built ML systems at Iguazio, Guesty, Enphase, Mila.
Summary

Read time: 6 min 13 sec

Top News

▸ OpenAI releases GPT-5.5 with faster speed and stronger agentic coding

AugmentCode

▸ Your prompts go stale. Your spec shouldn't.

Top News

▸ OpenAI launches workspace agents in ChatGPT to run team workflows across Slack and more

Top News

▸ Qwen releases open-source 27B model that beats its own 397B model on coding

Signals

1. Google open-sources its AI design guidelines days after Anthropic locked them down
2. Open-source tool turns any document into a fully editable PowerPoint deck
3. New recurrent model generates text embeddings with constant memory, beating transformer limits
4. Unsloth releases quantized Qwen3.6-27B with fine-tuning and tool calling support
5. MIT's recursive model framework handles 10 million tokens, scoring 58x better than standard models
Top News
OpenAI releases GPT-5.5 with faster speed and stronger agentic coding
53,484 Likes
OpenAI releases GPT-5.5 with faster speed and stro

OpenAI just dropped GPT-5.5, and the big idea is simple: stop babysitting your AI. You hand it a messy, multi-step task, and it plans, uses tools, checks its own work, and pushes through to the finish line.

The biggest wins are in coding, computer use, and research. Early testers cut debug time from days to hours. One math professor built an algebraic geometry app from a single prompt in 11 minutes.

Here's what makes it worth paying attention to:

  • 82.7% on Terminal-Bench 2.0, which tests real command-line workflows
  • 58.6% on SWE-Bench Pro for resolving actual GitHub issues
  • Same response speed as GPT-5.4, but uses fewer tokens to finish the same jobs
  • API pricing starts at $5 per million input tokens, with a 1-million token context window

It's live now in ChatGPT and Codex for Plus, Pro, Business, and Enterprise users. API access is coming very soon.

Read more
forward →
Presented by AugmentCode
Your prompts go stale. Your spec shouldn't.

You add agents to move faster, then spend that time re-explaining context and passing decisions between sessions. The more agents you run, the more you become the system holding them together. You're not the developer anymore, you're the prompt janitor.

Intent fixes that with a living spec. Write the goal once. Agents read it, update it as code changes, and hand off cleanly. No re-briefing. No context collapse.

Free with Claude Code, OpenCode, and Codex.

Build with intent
partner with us →
Top News
OpenAI launches workspace agents in ChatGPT to run team workflows across Slack and more
19,868 Likes
OpenAI launches workspace agents in ChatGPT to run

OpenAI just shipped Workspace Agents in ChatGPT, and it's basically the upgrade Custom GPTs never were. Think of it as building a bot once, then your whole team uses it forever.

Here's the core idea: teams can create shared agents that handle complex tasks and long-running workflows, within the permissions set by their organization. They run in the cloud, so they keep working even when you're not.

What can you actually automate? Things like:

  • Triaging software requests, enforcing policy, routing approvals, and opening IT tickets
  • Capturing feedback from Slack and support channels, then turning signals into weekly product action
  • Researching accounts, summarizing call transcripts, and posting summaries into Slack, cutting 5-6 hours of weekly work per rep

To get started, click Agents in the ChatGPT sidebar, describe a workflow your team does often, and ChatGPT guides you step by step to turn it into an agent.

Workspace agents are free until May 6, 2026, after which credit-based pricing begins. Available on Business, Enterprise, Edu, and Teachers plans only.

Read more
forward →
Top News
Qwen releases open-source 27B model that beats its own 397B model on coding
13,025 Likes
Qwen releases open-source 27B model that beats its

A 27B model just beat a 397B one on coding. That's the headline. Qwen3.6-27B is Alibaba's new open-source model, and it's smaller but somehow better at coding than its much bigger sibling.

Here's what the numbers look like:

  • SWE-bench Verified: 77.2 vs 76.2 for the 397B model
  • Terminal-Bench 2.0: 59.3 vs 52.5
  • SkillsBench: 48.2 vs 30.0
  • Runs locally on roughly 18GB VRAM, Apache 2.0 licensed

One cool new feature: Thinking Preservation. Normally, AI models forget their reasoning after each reply and start fresh. This model can remember its reasoning across a whole conversation, which makes multi-step coding tasks faster and less repetitive.

It also reads images and video natively, supports a 262K token context window (think: entire codebases at once), and works with vLLM and SGLang out of the box. You can grab it on Hugging Face right now.

Read more
forward →
Signals
1
Google open-sources its AI design guidelines days after Anthropic locked them down
5,528 Likes
2
exe.dev reimagines cloud dev with instant persistent VMs - SSH in, build anything, deploy without touching infra
Presented by exe.dev
3
Open-source tool turns any document into a fully editable PowerPoint deck
7,466 Stars
4
New recurrent model generates text embeddings with constant memory, beating transformer limits
Must Read
5
Unsloth releases quantized Qwen3.6-27B with fine-tuning and tool calling support
131,398 Downloads
6
MIT's recursive model framework handles 10 million tokens, scoring 58x better than standard models
5,428 Likes
At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 200,000 developers stay informed and ahead. We're passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.

Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We're based in the United States. By using our site, you agree to be governed by U.S. laws.
Looking to promote your company, product, service, or event to 250,000+ AI developers?
Work with us