Gemini 4 Carbon
AI overview
Sign in and the AI will write an overview from our coverage.
Headlines · 10
- Cognition benchmarks Gemini 4 Carbon and Argon on FrontierCode 2.0
Cognition reported that Gemini 4 Carbon scored 40% on its FrontierCode 2.0 coding benchmark, with a cost of $4.04 per task. Gemini 4 Argon scored 26.3% at $6.41 per task, ranking ahead of Muse Wonder 1.0 but behind Claude Haiku 5.5 and Grok 4.7.
cognition (X) · 🔥 3 - Gemini 4 Carbon Tops Intelligence Index, but Access Is Limited
The Rundown AI reports that Google unveiled Gemini 4 Carbon, a frontier model focused on cybersecurity and long-running coding work. The model scored 59 on Artificial Analysis’s Intelligence Index, taking the top position from Claude Opus 5.5, and led 15 of 17 benchmarks in Google’s charts. Its introductory API price is $2 per million input tokens and $10 per million output tokens, but access is limited to select cyber defenders and no wider release date was reported.
The Rundown AI · 🔥 0 - Gemini 4 Carbon reportedly leads AA Intelligence Index
TestingCatalog reports that Gemini 4 Carbon scored 59 on the AA Intelligence Index, placing it first in the cited ranking. The same brief says Google is rolling the model out first to cyber defenders, but it provides no launch date or direct announcement from Google.
testingcatalog (X) · 🔥 0 - Google Releases Gemini 4 Carbon for Coding and Cybersecurity
爱范儿 reports that Google DeepMind released Gemini 4 Carbon, a flagship model focused on long-horizon coding and cybersecurity. Artificial Analysis Intelligence Index v5.0 reportedly scored Carbon (High) at 59, ahead of Claude Opus 5.5 (Max) at 58, while Google reported results on SEC-bench Pro, ExploitGym and CWE-bench. Access is initially limited to selected cybersecurity defenders and trusted partners through the Fairwind Program, with broader availability unannounced.
爱范儿 · 🔥 0 - Gemini 4 Carbon Tops Text Arena and Ranks Second in Agent Arena
Arena.ai reported that Gemini 4 Carbon (High) debuted at No. 1 in Text Arena, No. 2 in Agent Arena, and No. 4 in Code Arena: WebDev. Its Agent Arena result was based on more than 21,900 real-world agentic sessions, with a reported 16.2% net improvement and a median cost of $1.80 per task. Arena.ai also said the model reached the Pareto frontier across all three arenas.
arena (X) · 🔥 0 - Google launches Gemini 4 Carbon for cybersecurity and long-horizon tasks
Google DeepMind announced Gemini 4 Carbon as its next frontier model, emphasizing cybersecurity, coding, and long-horizon capabilities. Artificial Analysis measured it at 59 on Intelligence Index v5, placing it first, though it trailed Claude Opus 5.5 on the Coding Agent Index. Access initially goes to selected cybersecurity defenders through the Fairwind Program, with no general-availability timetable announced.
Google DeepMind Blog · 🔥 19 - Kimi Teases K3.1 as Anonymous Arena Model Kibitz Draws Coding Tests
Kimi posted a 24-second teaser showing Kimi Code with “Model: Kimi K3.1” displayed, but did not announce a release date, parameter count, or pricing. An anonymous Arena model named Kibitz has produced frontend and 3D coding demonstrations. Community observers inferred that Kibitz may be Kimi K3.1, partly based on tokenizer characteristics, but this identity remains unconfirmed. Informal comparisons were mixed: Kibitz completed one voxel-courtyard test faster than GPT-6 Astra Max, while another tester favored Gemini 4 Carbon over Kibitz, Kalmia, and Claude Opus 5.5.
AIbase AI新闻 · 🔥 0 - Gemini 4 Carbon reportedly enters partner early access and LM Arena testing
Reports say Gemini 4 Carbon has entered an invite-only partner early-access program and is being tested on LM Arena, but it has not been formally released. Testers described Carbon as noticeably better than Gemini 4 Argon at 3D generation, including a complete Super Mario level, while reporting that it was roughly two to four times slower depending on the task. They also described its SVG and frontend/UI output as weak. The comparisons come from arena matchups and hand-picked prompts rather than standardized benchmarks, and the checkpoint served on LM Arena cannot be independently verified.
DoNews · 🔥 0 - OpenAI reportedly prepares GPT-6.1 Terra ahead of upcoming release
TestingCatalog, citing Lumina, reports that Microsoft renamed 18 evaluation configs from GPT-5.6 Terra to GPT-6.1 Terra, a change that typically happens when a team is preparing to test a model. The report lines up with The Information's story, shared by Stephanie Palazzolo, who said this month should be busy for model releases including a new Anthropic model, Google's Gemini 4 Carbon and "a potential new Terra model from OpenAI." The new Terra is reported to beat GPT-6.1 Sol at a lower price, with Sol currently priced at $2 per million input tokens and $10 per million output tokens, though Palazzolo later said some launches could slip into November.
36氪 人工智能 · 🔥 0 - Google staff reportedly test Gemini 4 Carbon, compare coding to Opus 5.5
Business Insider reports that Google staff are testing a new Gemini 4 version codenamed Carbon on the internal coding platform Jetski. One employee said it “feels like Opus 5.5” for coding but cautioned that more testing is needed; Google has tested Argon, Barium and Carbon variants, and it is unclear whether Carbon will ship as an Argon update or a separate model. Google declined to comment, while the already-announced Argon is not yet broadly available and Antigravity has shown references and subscription prompts for it.
TestingCatalog · 🔥 0
Experience and discussion from the community
Share my Gemini 4 Carbon experienceAsk about Gemini 4 Carbon
Nobody has shared their experience with Gemini 4 Carbon yet.