Models

Gemini 4 Hydrogen: price, benchmarks and Antigravity's new Orchestrator; Claude Code adds Code Mode, Codex CLI 0.168.0

Google released Gemini 4 Hydrogen on November 12, a fast model for everyday coding and agents at $1.50/$7.50 per million tokens, with a free API tier and a new Orchestrator in Antigravity that pairs it with Carbon or Argon. Also: Code Mode in Claude Code 2.1.323, Codex CLI 0.168.0, the agent-built AWS Emulator and Nebius's $40-45 billion capex plan for 2027.

By ·

Google released Gemini 4 Hydrogen on November 12, 2026. It is the fast model of the Gemini 4 family, built for everyday coding and agent work, next to the frontier models Gemini 4 Carbon and Gemini 4 Argon (Google announcement). Google calls it the recommended upgrade from Gemini 3.8 Flash.

Gemini 4 Hydrogen: price and specs

Gemini 4 Hydrogen: price and specs
ItemGemini 4 Hydrogen
Model IDgemini-4-hydrogen
API price (paid tier)$1.50 per million input tokens, $7.50 per million output tokens (thinking included), flat across the whole context window
Cached input$0.075 per million tokens, implicit caching only (no explicit caching)
Free tierYes, rate-limited, in the Gemini API and Google AI Studio
ContextOne 1,048,576-token window shared by input, thinking and output
OutputUp to 65,536 tokens per request; up to 1M with background execution
Thinking levelslow, medium, high (default) and a new xhigh

Figures from the Gemini API guide for Hydrogen and the Gemini API pricing page. For comparison, Gemini 3.8 Flash costs $0.75/$3.75 until December 31, 2026, then $1.50/$7.50 from January 1, 2027, the same list price as Hydrogen (same pricing page). Migration notes from Google: Hydrogen rejects the minimal thinking level, ignores temperature, top_p and top_k, and does not support explicit context caching.

Benchmarks (Google figures)

Google says Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default high level, ahead of Argon's 50, and 54 at xhigh. Against Gemini 3.8 Flash, it reports 57% vs 18.2% on Terminal-Bench 4.1, tasks finished 2.5 times faster at the default level and more than 5 times fewer fabricated claims. Other scores given by Google: 58.8% on HLE-Diamond and 52.3% on Vals Tax Agent Bench (Google announcement). These are vendor figures, not independent tests, and we have not tested the model.

Where you can use it

On the morning of November 13, GitHub's changelog listed no Hydrogen release for Copilot (GitHub changelog).

Orchestrator: Carbon plans, Hydrogen executes

Google also added Orchestrator to Antigravity's model picker. A planner model (Gemini 4 Carbon or Argon) explores the code and writes a plan in a read-only phase, Hydrogen then edits files and runs builds and tests, and control goes back to the planner when Hydrogen hits friction, for example a check that fails twice (Antigravity docs). Orchestrator (Carbon) and Orchestrator (Argon) are rolling out to Google AI Ultra; paid Google AI Pro gets Orchestrator (Argon) (Antigravity blog).

According to Google, Orchestrator reached near-Carbon task success while using 24% less quota and finishing 19% faster on average. The Antigravity blog places that test on Terminal-Bench 4.1, the main Google post on Terminal-Bench 4.0: the version is not consistent between the two pages.

Claude Code 2.1.323: Code Mode

Anthropic released Claude Code v2.1.323 on November 12 (22:12 UTC). The main addition is Code Mode, an experimental opt-in mode in which Claude writes JavaScript to call its tools; organizations can turn it off. The release also adds an engines["claude-code"] field to plugin.json (a plugin outside its stated version range is not installed, updated or loaded), .docx uploads in the Artifact tool, and a long list of permission fixes: for example, --permission-mode auto now starts in Manual mode when auto mode is unavailable (release notes). Our Claude Code page now shows this version.

Codex CLI 0.168.0

OpenAI released Codex CLI 0.168.0 on November 13 (03:54 UTC) (release notes):

Our Codex page now shows this version.

AWS Emulator, written by agents

On November 11, AWS released AWS Emulator, a local emulator of AWS APIs that the AWS CLI, the SDKs and coding agents reach over a local HTTPS endpoint. It is generally available with no additional service fee, written in Rust, installed from RPM or Debian packages on Linux or run in Docker on macOS and Windows (AWS News Blog). AWS says autonomous agents wrote all of its code and that its team did not write or review any of it; the source is on GitHub under CC0, except the vendor/ folder under Apache-2.0 (About Amazon, GitHub repository). Limits at launch: no IAM policy evaluation as on AWS, one account per emulator, and fidelity that varies by service. To let an agent set it up (AWS CLI 2.35.0 or later):

aws agent-toolkit add-skill --skill-name aws-emulator --region us-east-1

Infrastructure: Nebius plans $40-45 billion of capex in 2027

Nebius reported third-quarter revenue of $980.8 million, up 571% year over year (Nebius results). In its shareholder letter, the company says customers have committed more than $12 billion in contract value since the start of the third quarter, now expects 2026 revenue at the low end of its $3.0-3.4 billion range and capex of about $27 billion (above the earlier $20-25 billion), and gives a first 2027 outlook: $12-13 billion in revenue and $40-45 billion in capex (shareholder letter). These are company forecasts. The shares fell 16% on November 12, according to CNBC.

Sources

Spotted an error in this article? Report it · Our corrections policy

OrigoTech logo

New: the app iPhone & iPad

OrigoTech is now an iPhone and iPad app

Read the articles natively, with “New” markers on unread ones, local search and offline reading of what you have already loaded. Free, no account, iOS 17 or later; ads appear between articles.

Rekha icon: a house drawn in a single stroke

From the same author Self-promotion

An Eulerian path, drawn with your finger

Rekha, the iPhone game by OrigoTech's author: connect every dot in a single stroke without tracing the same link twice, across 500 drawings that are all solvable.

The OrigoTech newsletter

A sourced weekly digest of AI tool news.

Read next

Agents

Antigravity Dynamic Workflows: /run-workflow turns agent teams into Python scripts; Copilot drops six models on December 14, Codex CLI 0.169.0

Antigravity 2.32.0 (Nov 13) adds Dynamic Workflows, where the agent writes a short Python script that runs many subagents in parallel or in sequence, and you approve the script before it runs. GitHub put Gemini 4 Hydrogen in Copilot and will retire six Copilot models on December 14. Also: Codex CLI 0.169.0, Claude Code 2.1.325, Zed 1.28 and AirTrunk's 1.8 GW plan in India.

· 6 min read