---
title: 'In the News: November 12, 2026 (Extra 2)'
description: "Google's Antigravity adds Orchestrator, which passes one coding session between Gemini 4 Carbon and the cheaper Gemini 4 Hydrogen released today."
canonical_url: 'https://darkfactory.dev/news/2026-11-12-extra-2'
markdown_url: 'https://darkfactory.dev/news/2026-11-12-extra-2.md'
collection: news
date_published: '2026-11-12T16:10:00-05:00'
date_modified: '2026-11-12T16:10:00-05:00'
---

# In the News: November 12, 2026 (Extra 2)


Google released Gemini 4 Hydrogen today at $1.50 per million input tokens and $7.50 per million output tokens. Its Antigravity coding tool also gained a mode called Orchestrator, which passes a single session back and forth between Hydrogen and the larger Gemini 4 Carbon. Separately, Parcha open-sourced AgentRun and published the numbers behind a code reviewer it built on a cheap model after its review bill rose from $180 to $6,451 a month.

## 1. Antigravity's Orchestrator switches between a planning model and a cheaper executing model inside one session

**[Gemini 4 Hydrogen and Orchestrator](https://antigravity.google/blog/gemini-4-hydrogen-in-google-antigravity/)** · The Antigravity Team, Google · Google Antigravity Blog, November 12, 2026

Orchestrator is a new option in Antigravity's model selector. It pairs Gemini 4 Carbon with Gemini 4 Hydrogen, a model Google [introduced the same day](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-hydrogen/). The Antigravity post sets it against a static "model router" that "guesses task difficulty from your first prompt" and sends the whole task to one model. In the post's example, Carbon starts in a read-only phase, explores dependencies and builds a plan "with explicit re-planning triggers". Hydrogen then writes the edits, updates packages and runs tests. If Hydrogen "hits friction", it hands back to Carbon, which inspects the state, repairs the plan and resumes execution.

Google reports that Orchestrator reached "near-Carbon task success rates" while using 24% less quota and finishing 19% faster on average. Neither post gives the success rates themselves, and the two posts name different benchmark versions for the same result: Terminal-Bench 4.1 in the Antigravity post, Terminal-Bench 4.0 in the model announcement. For Hydrogen by itself, Google lists a flat price of $1.50 per million input tokens and $7.50 per million output tokens across a 1 million-token context window, and a score of 57% on Terminal-Bench 4.1 against 18.2% for Gemini 3.8 Flash. Hydrogen has four thinking levels: low, medium, high (the default) and a new xhigh. Every figure here is Google's own.

In Antigravity, Hydrogen is rolling out to paid Google AI Pro and Ultra subscribers. Orchestrator is rolling out to Ultra subscribers, with more models and plans to follow, and Google says enterprises can reach both through Gemini for business. For teams that want to assign models to roles themselves, the post points to the Antigravity SDK and shows an issue-triage demo in which Hydrogen agents triage and code while a Carbon agent takes larger refactors, new features and security checks.

**Why it matters:** A router that picks a model from the first prompt has to judge difficulty before the work starts. Orchestrator makes that choice during the run and lets the cheaper model hand the task back. If you run your own harness, the design the post describes is a planner that writes down when to re-plan and an executor that returns control when it gets stuck. The 24% quota saving comes from a benchmark for which Google does not report success rates, so measure cost per completed task on your own work before you count on it.

## 2. Parcha says a read-only shell took its cheap-model code reviewer from wrong 70% of the time to 7%

**[AgentRun (README)](https://github.com/Parcha-ai/agentrun/blob/main/README.md)** · Parcha Labs · GitHub, Apache-2.0; posted to Hacker News November 12, 2026

AgentRun is an open-source control plane that sits on top of an existing harness: Claude Code, Codex, OpenCode or pi. It adds durable runs, approvals, spend caps and a learning loop. With learning turned on, the agent gets one more turn at the end of each run to say what it would keep: a note, a skill or a workflow. AgentRun grades the proposals and commits the keepers to the agent's own git folder. Once a learned skill has served three runs, the next run is asked to draft that job as a workflow, and the draft is dry-run before it is kept. The README's own line is "A lot of agents don't need to be agents."

The numbers are in the [code review case study](https://github.com/Parcha-ai/agentrun/blob/main/docs/case-studies/code-review.md). Parcha says its bill from Greptile, a code review product, went from $180 a month to $6,451 after Greptile moved in March to a dollar for every review past 50 and Parcha's own agents "took off". It built a replacement on a model it names particle-1. One parent agent sends out eight reviewers on a read-only checkout of the pull request, five quick and three deep, checks every finding against the code and posts one review. Parcha puts the cost at about 18 cents a review. It tuned on 70 merged PRs and scored on the 30 newest PRs in the AgentRun repository, which it had kept aside. On serious bugs its reviewer caught 5 of 8 and Greptile caught 1; the README gives the overall result as 1.6 times the real bugs Greptile caught.

Two of its lessons carry over to any reviewer. The first: "Reading only the diff, the cheap model was wrong 70% of the time. With read-only bash on the code, 7%." The second concerns scoring: a timeout that kept reviews from posting once made a good version look 15 points worse.

This is a vendor measuring its own product on its own repository. The answer key of 78 bugs, 8 of them serious, was built from findings graded by Opus 5.5, with no human grading described, and eight serious bugs is a small base. The write-up does not define "wrong". We read the README and the case study. We did not run AgentRun or open its tuning tutorial. The Hacker News post had 48 points and 5 comments about 2.8 hours after it went up.

**Why it matters:** Parcha's review bill rose 36 times in a year under per-review pricing, and this is its account of replacing the paid reviewer with a tuned cheap model. Two details are cheap to try on your own reviewer: give it read-only shell access to the checkout instead of the diff alone, and count the reviews that never posted when you score it. The comparison with Greptile rests on 30 PRs and an answer key graded by a model, so read the 5-of-8 figure as Parcha's claim.