Gemini 4 Carbon
blog.google ·
Google has announced Gemini 4 Carbon, a frontier AI model focused on long-horizon engineering, coding and cybersecurity defense. Google says Carbon leads version 5.0 of the Artificial Analysis Intelligence Index with a score of 59 and will initially be available this week to a subset of cyber defenders, with broader access phased through its Fairwind Program. Google is participating in the U.S. government’s voluntary pre-release model-testing process before extending access to developers, businesses and consumers.
Carbon is priced at $2 per million input tokens and $10 per million output tokens, matching Gemini 4 Argon’s introductory pricing; cached input tokens receive a 95% discount. Google says the model can sustain agentic engineering workflows for days, including diagnosing and patching subtle defects in large codebases. It scored 88.2% on SEC-bench Pro for vulnerability discovery, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching, all evaluated without guardrails.
Google reports that Carbon reduced high-severity misalignment alerts by almost 80% in internal live coding environments, cut unsupported claims by about half relative to Argon, and achieved a 0.0% attack-success rate on Gray Swan’s Indirect Prompt Injection benchmark. It also scored 1768 on WebDev Arena, ranking fourth after an 89-point increase.
What people are saying
The thread spent far less time celebrating Gemini 4 Carbon than trying to work out what it is, who can use it, and whether its benchmark deck means anything before release.
What readers liked
- Carbon’s launch pricing promises frontier-class input and output costs that could make heavy AI use feasible without upgrading to expensive premium…
- Users familiar with Argon describe it as a strong model for its price, making a materially better Carbon at the same rate look like exceptional value.
Concerns raised
- Element-themed names obscure whether Carbon supersedes Argon or occupies another capability tier, making model selection needlessly opaque for…
- Publishing benchmark results before broad access feels hollow when competitors can advance before anyone can verify the claimed advantage in…
Read full discussion analysis
Source: Hacker News · Based on 15 sampled comments · Analyzed
Discussion overview
The thread spent far less time celebrating Gemini 4 Carbon than trying to work out what it is, who can use it, and whether its benchmark deck means anything before release. Google’s element names leave readers unable to tell whether Carbon replaces Argon, sits above it, or is simply a revised version at the same price. That ambiguity is especially irritating because many people still cannot find the recently announced models in the normal Gemini product, or are gated by subscription tier.
There is real interest beneath the annoyance. Readers who have used Argon call it capable and unusually good value, while Carbon’s introductory token pricing looks aggressively cheap against premium competitors. Reported lower token use than Opus and stronger coding results make the prospect of a same-price upgrade compelling. One technical claim drew particular approval: reducing benchmark shortcutting while raising coding accuracy addresses a failure mode that made prior Gemini models hard to trust in agentic work.
But commenters largely reject a pre-release benchmark victory lap as proof of leadership. A fast-moving market can erase an announced advantage before customers receive it, and restricted early access makes the announcement feel like marketing rather than a usable launch. Several also doubt that raw scores alone can dislodge Claude in coding, where trust, established workflows, and internal adoption still matter. Others counter that model loyalty is shallow when a clearly better or cheaper option appears.
Points of disagreement
The sharpest split is whether Carbon’s announcement establishes a meaningful frontier advance now or is empty pre-release positioning. Supporters point to excellent claimed value, improved benchmark performance, token efficiency, and anti-cheating progress; skeptics argue that unavailable models and delayed access make scores perishable, especially against already accessible rivals.
What readers liked
- Carbon’s launch pricing promises frontier-class input and output costs that could make heavy AI use feasible without upgrading to expensive premium plans.
- Users familiar with Argon describe it as a strong model for its price, making a materially better Carbon at the same rate look like exceptional value.
- Reported reductions in benchmark cheating alongside higher coding accuracy target a concrete reliability flaw that undermines autonomous coding agents.
- Lower token consumption than a premium rival, if sustained in real tasks, could turn comparable reasoning into a substantially cheaper workflow.
Concerns raised
- Element-themed names obscure whether Carbon supersedes Argon or occupies another capability tier, making model selection needlessly opaque for developers.
- Publishing benchmark results before broad access feels hollow when competitors can advance before anyone can verify the claimed advantage in production.
- Subscription gating and delayed visibility in Gemini’s own interface make the rollout feel exclusionary and poorly coordinated.
- Benchmark strength alone may not win coding users from Claude, whose trust, ecosystem, and entrenched workflows outweigh a fresh scorecard.
Keep reading
Sources and details
blog.google · Published
Gemini 4 Carbon
Read report →
Community analysis available for 1 of 1 linked discussions.
Additional discussion searches are still incomplete.
Coverage details and dates
Latest available community analysis:
Summaries: 1 current · 0 awaiting update · 0 not verified
Latest summary update:
Hacker News · blog.google