Google Unveils Gemini 4 Carbon AI Model But Keeps It Under Wraps With Restricted Fairwind Access

Google Claims Top Benchmark Position With Gemini 4 Carbon

Google announced Gemini 4 Carbon on October 28, positioning it as the world's most capable frontier AI model just two weeks after releasing Gemini 4 Argon1

1

. Independent evaluator Artificial Analysis places Carbon first on its Intelligence Index v5.0 with 59 points, ahead of Claude Opus 5.5 at 58 points, GPT-6 Astra at 57 points, and Argon at 50 points2

2

. Carbon achieves the highest results in 15 of 17 benchmark rows in Google's announcement, though OpenAI retains an advantage on one software engineering test and Anthropic leads on one computer-use evaluation2

2

. The model represents an incremental improvement over Gemini 4 Argon, beating it by slim margins in every test Google conducted1

1

.

Cybersecurity Tasks Drive Carbon's Development Focus

Google designed Gemini 4 Carbon specifically for cybersecurity tasks including vulnerability discovery, exploitation, and software patching3

3

. The AI model demonstrates a significant leap in vulnerability detection, validation, and remediation1

1

. Carbon generates secure code at more than twice the success rate of Gemini 4 Argon on internal benchmarks3

3

. The Chrome team reports that Carbon performs 10 times better at generating high-quality patches compared to non-Google models1

1

. In the Grey Swan evaluation measuring resistance to prompt-injection attacks, Carbon achieved a 0 percent score, outperforming Argon at 0.7 percent, Claude Opus 5.5 at 1 percent, and GPT-6 Astra at 8.5 percent1

1

.

Complex Software Engineering Capabilities Extended

Carbon handles complex, long-running engineering and coding tasks with improved reliability3

3

. Google's internal teams report seeing Carbon run for days on sustained agentic tasks, continuing through unsuccessful approaches while diagnosing bugs and developing patches in substantial codebases2

2

. The model increased its score by 89 points on WebDev Arena, now ranking at number four on the leaderboard1

1

. Independent testing reveals tradeoffs in Carbon's performance: while it improves substantially over Argon in coding and knowledge work, it trails several rivals on professional tasks and records a higher hallucination rate than Argon on factual-knowledge tests2

2

.

Google's Fairwind Program Controls Initial Rollout

Google is making Gemini 4 Carbon available first only to a subset of cybersecurity defenders through its Fairwind Program as the company continues testing safeguards3

3

. The company is working within the voluntary testing framework established by the US government and plans to roll the model out through its Fairwind cyber defense program once testing completes1

1

. Google says broad release will come as soon as possible after gathering feedback and iterating on guardrails2

2

. The release path mirrors Gemini 4 Argon's staged approach, which Google announced on September 30 and opened more broadly on October 132

2

. Carbon will likely arrive first for AI Ultra subscribers and developers using the Gemini API, though Google hasn't provided a timeline1

1

.

Token Prices Match Argon But Timeline Remains Unclear

Google will price Gemini 4 Carbon at $2 per million input tokens and $10 per million output tokens, matching Gemini 4 Argon's introductory pricing3

3

. The model offers a 95 percent discount for cached inputs2

2

. These rates are temporary, with subsequent prices set at $4 per million input tokens and $20 per million output tokens2

2

. Argon's discounted rates expire at the end of November, but Google hasn't specified when Carbon's introductory period will end1

1

. The announcement doesn't clarify whether Carbon shares Argon's model architecture or parameter count, nor does it provide an equivalent API specification establishing supported features and context limits2

2

.