This Week in AI: OpenCode Hits #1, OpenAI's $1T IPO, and DiffusionGemma

A lot happened in AI this week. The AI coding tool rankings changed. Two big AI companies filed to go public. And Google released a model that writes text in a new way.

Here is everything that matters for developers, in one short read.

1. OpenCode Is the New #1 AI Coding Tool

By GitHub stars, OpenCode now leads every open-source AI coding tool.

But popularity is not the full story. OpenCode has one feature the others do not.

It is the only major tool (besides Lovable) with a true offline, air-gapped mode. It can run local models through Ollama. When you use a local model, your code never leaves your machine.

This matters for teams with strict data rules — defense, healthcare, and finance. They often cannot send code to a cloud API. OpenCode gives them a real option.

The first step is running a coding model locally with Ollama:

# Pull and run a local coding model — this stays on your machine
ollama run qwen3-coder

Then set Ollama as your provider in OpenCode’s config and point it at that local model. Check the OpenCode docs for the exact settings, since the config format changes between versions.

2. But Stars Are Not Skill

GitHub stars measure popularity, not coding ability. On Terminal-Bench — a test of real command-line task completion — the ranking flips.

So the most-starred tool is not the strongest one. Pick your tool for what it does well, not for its star count.

3. GitHub Copilot and Windsurf Both Changed

Two more shifts landed this week.

GitHub Copilot moved to usage-based credits on June 1. One credit equals one cent ($0.01). GitHub also pushed deeper into multi-agent tooling — you can now coordinate several coding agents at once, instead of one assistant at a time.

Windsurf rebranded. Reports say it relaunched under the Devin Desktop name, tying it to Cognition’s Devin agent.

4. OpenAI and Anthropic Are Going Public

The companies behind your AI tools are heading to the stock market.

On June 8, OpenAI filed a confidential S-1 (the first step toward an IPO). Its announcement was blunt:

“We expect it to leak, so we’re just announcing it.”

OpenAI was last valued at about $852 billion (March 2026). Reports say it is now targeting a listing near $1 trillion that could raise $60 billion or more.

Anthropic filed about a week earlier. It reportedly closed a round near $965 billion — briefly the more valuable of the two.

Why does this matter to developers? Once these companies answer to public shareholders, their choices change. Pricing and model access stop being about research. They start being about revenue. Watch your API costs.

5. Google’s DiffusionGemma Writes Text Differently

Google released DiffusionGemma on June 10. It throws out a core idea about how language models work.

Normal models write one token at a time, left to right. DiffusionGemma uses diffusion instead — the same idea behind image generators. It refines a whole 256-token block at once, then sharpens it over a few passes.

The result is speed:

• Up to 4x faster text generation
• Over 1,000 tokens per second on a single H100 GPU
• A 26B mixture-of-experts model, but only about 4B parameters run at once
• Runs in roughly 18 GB of memory — consumer-grade hardware
• Open weights, Apache 2.0 license

This is the first real look at what “post-autoregressive” models might feel like. You can download and run it today.

Quick Hit: Microsoft Web IQ

Microsoft launched Web IQ — a set of APIs that let AI agents query the Bing index directly. Instead of raw web pages, it returns selected passages and structured evidence. The goal is to cut the tokens and latency that agents waste on web search. If you build agents that browse the web, this one is worth a look.

The Real Trend

Step back and one theme is clear. Models are getting cheaper and faster. Open weights are catching the frontier. And the real competition is moving from the model to the tooling around it.

In other words: the moat is the workflow now. The model you use matters less each month. The tool you wrap around it matters more.

Related Articles

• Claude Opus 4.8: 5 Things That Changed for Developers
• Build Your First AI Coding Agent

Follow @kemal_codes on X and kemalcodes on GitHub. New AI news roundup every week.