Google unveils Gemini 4 Carbon AI model, pitched at cybersecurity and coding
Google unveiled Gemini 4 Carbon on 28 October, two weeks after Gemini 4 Argon. It says the model leads on cybersecurity benchmarks. Access starts with a group of security defenders.
Updated (version 2). New coverage since the last version from Ars Technica, The Verge.
1 / 4
The story, neutrally told
Centre · 2Google unveiled Gemini 4 Carbon, which it calls its latest frontier AI model, on Wednesday 28 October 2026. CNAN “Alphabet's Google on Wednesday unveiled Gemini 4 Carbon, its latest frontier AI model” Read at CNA ↗ Anadolu AgencyN “Google announced its latest artificial intelligence (AI) model, Gemini 4 Carbon, on Wednesday” Read at Anadolu Agency ↗ Centre · 1The launch came two weeks after Google introduced Gemini 4 Argon. CNA described it as stepping up competition with OpenAI and Anthropic in coding and cybersecurity AI. CNAN “The launch comes two weeks after Google introduced Gemini 4 Argon”“stepping up competition with OpenAI and Anthropic” Read at CNA ↗ Centre · 1Google attributed the quick release timeline to its researchers and AI models "working hand in hand to advance capabilities faster". Anadolu AgencyN “working hand in hand to advance capabilities faster” Read at Anadolu Agency ↗
Centre · 1Google said Carbon delivers state-of-the-art cybersecurity performance, including vulnerability discovery, exploitation and software patching. It said the model is designed for complex, long-running engineering and coding tasks. CNAN “state-of-the-art performance across cybersecurity tasks including vulnerability discovery, exploitation, and software patching”“designed to handle complex, long-running engineering and coding tasks” Read at CNA ↗ Centre · 1Per Anadolu Agency, Google reported scores of 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation and 71% on CWE-bench for patching. It said it has a "firm lead" on these benchmarks. Anadolu AgencyN “scoring 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation and 71% on CWE-bench for patching” Read at Anadolu Agency ↗ Centre · 1Google said that on internal benchmarks Carbon generates secure code at more than twice the success rate of Gemini 4 Argon. CNAN “Carbon generates secure code at more than twice the success rate of Gemini 4 Argon on internal benchmarks” Read at CNA ↗
Centre · 1Google also claimed Carbon ranks first on version 5.0 of the Artificial Analysis Intelligence Index with a score of 59. It claimed an 89-point jump to fourth place on the WebDev Arena leaderboard. These are the company's own claims as relayed by Anadolu. Anadolu AgencyN “ranks first on the updated version 5.0 of the Artificial Analysis Intelligence Index with a score of 59”“claiming an 89-point jump to fourth place on the WebDev Arena leaderboard” Read at Anadolu Agency ↗ Left · 1The Verge, which also relayed Google's announcement, reported that Carbon holds the top spot on the Artificial Analysis Intelligence Index (v5.0), with Claude Opus 5.5 and GPT-6 Astra trailing behind, and described "significant gains" in coding, agentic tasks and cybersecurity. The VergeLC “Google highlights Carbon's top spot on the Artificial Analysis Intelligence Index (v5.0) benchmark, with Claude Opus 5.5 and GPT-6 Astra trailing behind.” Read at The Verge ↗ Centre · 1On safety and alignment, Google said internal testing showed Carbon cuts unsupported claims by about half compared with Argon. It also said the model reduces task overreach by 26% and under-service by 29% in multi-step workflows. Anadolu AgencyN “Carbon cuts unsupported claims by approximately half compared to Argon”“reducing task overreach by 26% and under-service by 29% during multi-step workflows” Read at Anadolu Agency ↗
Centre · 2Access is limited at first. Carbon goes first to a subset of cybersecurity defenders through Google's Fairwind Program while the company continues testing safeguards. Anadolu reports that a broader rollout to developers, businesses and consumers will follow. CNAN “made available first to a subset of cybersecurity defenders through Google's Fairwind Program as the company continues testing safeguards” Read at CNA ↗ Anadolu AgencyN “before rolling out more broadly to developers, businesses and consumers” Read at Anadolu Agency ↗ Centre · 2Introductory pricing is $2 per million input tokens and $10 per million output tokens, matching Argon's introductory pricing. Anadolu adds that cached input tokens are discounted by 95%, and that standard pricing will later rise to $4 per million input tokens and $20 per million output tokens. CNAN “priced at $2 per million input tokens and $10 per million output tokens, matching Gemini 4 Argon's introductory pricing” Read at CNA ↗ Anadolu AgencyN “standard pricing rises to $4 per 1 million input tokens and $20 per 1 million output tokens” Read at Anadolu Agency ↗
Every sentence links to the reporting it rests on. The pill in front of each says where its sources sit: Left, Centre or Right when one side supplies at least half of them, Mixed when they are evenly split. The number is how many outlets it cites.
Left2 outlets
- Framing
- The Verge presents the launch as a product release, leading with Google's own "significant leap" in cybersecurity language and the head-to-head ranking against rival models.
- Emphasis
- Carbon's top spot on the Artificial Analysis Intelligence Index, with named rivals Claude Opus 5.5 and GPT-6 Astra behind it.
- Leaves out or plays down
- Pricing, the Fairwind Program's limited first access and the safety figures are not in the opening text.
- Charged language
- “significant leap”
- For example
-
“Google launches Gemini 4 Carbon with a ‘significant leap’ in cybersecurity” — The Verge
Centre2 outlets
- Framing
- Both centre outlets report the launch as a company announcement. CNA gives a short competitive-race summary with bullet points. Anadolu gives a detailed account of Google's benchmark claims, attributed to the company.
- Emphasis
- CNA stresses competition with OpenAI and Anthropic and the limited initial access. Anadolu stresses benchmark scores, safety claims and the pricing schedule.
- Leaves out or plays down
- Neither reports independent verification of the benchmark results. CNA omits the specific benchmark scores and the later price rise. Anadolu omits the competitor context.
- For example
-
“stepping up competition with OpenAI and Anthropic as the race to build more capable coding and cybersecurity systems intensifies” — CNA
“the tech giant claimed the model holds a "firm lead," scoring 88.2% on SEC-bench Pro for finding vulnerabilities” — Anadolu Agency
Right0 outlets
No right outlet in our sources has covered this story yet.
What every side reports
- Google unveiled Gemini 4 Carbon on 28 October 2026, two weeks after Gemini 4 Argon.
- Carbon is pitched at cybersecurity and long-running coding tasks.
- It is available first to cybersecurity defenders.
- Introductory pricing is $2 per million input tokens and $10 per million output tokens.
-
Gemini 4 Carbon delivers state-of-the-art cybersecurity performance.
Reported- Reports 3
- The Verge, Anadolu Agency, CNA
-
Google released Gemini 4 Carbon, its latest frontier AI model, on 28 October 2026
Supported- Supports 3
- The Verge, Anadolu Agency, CNA
Google organisation
Google presents Carbon as its strongest frontier model, with leading cybersecurity and long-horizon abilities. It is releasing it first to defenders while it continues testing safeguards.
“its strongest frontier model with state-of-the-art cybersecurity and long-horizon capabilities” — Anadolu Agency
“as the company continues testing safeguards” — CNA
Left2 articles
-
-
Google launches Gemini 4 Carbon with a ‘significant leap’ in cybersecurity
Neutral Reports the release largely in Google's terms, naming rival models trailing on the intelligence index.
Centre2 articles
-
CNAN ·
Google launches Gemini 4 Carbon AI model with cybersecurity focus
Neutral Brief wire-style report that places the launch in the competition with OpenAI and Anthropic and lists Google's claims as bullets.
-
US tech giant Google unveils Gemini 4 Carbon AI model
Neutral Detailed relay of Google's blog claims on benchmarks, safety and pricing, with each claim attributed to the company.
Right0 articles
No coverage yet.
- 28 Oct 18:08 First CNAN Google launches Gemini 4 Carbon AI model with cybersecurity focus
- 28 Oct 18:26 +18m Anadolu AgencyN US tech giant Google unveils Gemini 4 Carbon AI model
- 28 Oct 18:47 +39m Ars TechnicaLC Google announces Gemini 4 Carbon, its latest frontier AI model that you can't use
- 28 Oct 20:03 +1h 54m The VergeLC Google launches Gemini 4 Carbon with a ‘significant leap’ in cybersecurity
Times are when each article was published, or when we first saw it if the outlet gave no time.
Ask about this story
Loading…