Skip to content
ENTITY PROFILE / KNOWLEDGE GRAPH

Gemini 4 Carbon

RSS

Google DeepMind released Gemini 4 Carbon two weeks after Argon. It tops the Artificial Analysis Intelligence Index v5.0 with 59 points, surpassing former leaders Claude Opus 5.5 and GPT‑6 Astra. On WebDev Arena it ranks fourth (1,769 Elo), up 89 Elo from Argon. Carbon adds full cyber‑defence capabilities—detect, validate and patch vulnerabilities—and is initially limited to selected security experts. Benchmark results: 70.36 % accuracy on Vals AI, 68‑70 % overall on agentic tasks, beating Argon in all but Claude’s agentic coding. It blocks 0 % prompt‑injection in the Gray Swan test, raises answer accuracy from 48 % to 58 %, and produces 64 % more output tokens. Carbon achieves 89 % higher WebDev scores and a $1.80 task cost, 72 % below Claude Opus 5.5. It outperforms GPT‑6 in most tests, including ExploitGym, and is priced the same as Argon ($2 /10 M tokens). Google released Gemini 4 Carbon Gemini 4 Carbon offers significant gains over Argon in coding, agentic tasks, and cybersecurity Gemini 4 Carbon holds the number one spot on ExploitGym Gemini 4 Carbon is currently only available to a select group of cybersecurity defense companies in the Fairwind Program Fairwind Program provides access to Gemini 4 Carbon Gemini 4 Carbon ranked #1 on HLE-Diamond Gemini 4 Carbon ranked #1 on TB-Science Gemini 4 Carbon ranked #1 on AA-Omniscience Gemini 4 Carbon ranked #1 on AutomationBench Gemini 4 Carbon ranked #1 on Vending-Bench 2 Gemini 4 Carbon ranked #1 on Blueprint-Bench 2 Gemini 4 Carbon compared with Claude Code on Coding Agent Index (score 59 vs 62) Gemini 4 Carbon compared with Opus 5.5 on FrontierCode v1.1 (score 51% vs 56%) Gemini 4 Carbon compared with GPT-6 Astra on FrontierSWE (score 61.2% vs 65.5%) Carbon model was built by Google· Entity summary generated by Graphiti

RELATIONS
53Relations
CURRENT
45Current
EVENTS
2Events
REPORTS
7Source reports
KNOWLEDGE ATLAS0 entities · 0 relations
No public neighbors yet
Explore all published links of a node; previously discovered relations remain visible.

Relation timeline

BY VALID-FROM DATE
  1. Gemini 4 CarbonBUILT BYGoogle

    Carbon model was built by Google

    Current1 sourceGoogle
  2. DeepMindANNOUNCEDGemini 4 Carbon

    DeepMind ANNOUNCED Gemini 4 Carbon

    Current2 sourcesDeepMind
  3. Gemini 4 CarbonRANKED TOP ONBloombergNEF

    Gemini 4 Carbon ranked #1 on Vending-Bench 2

    Current1 sourceBloombergNEF
  4. Gemini 4 CarbonRANKS ATAA Intelligence Index v5

    Gemini 4 Carbon RANKS_AT AA Intelligence Index v5 score 59; #1 among 9 index evals

  5. Gemini 4 CarbonRANKED TOP ONAA-Omniscience

    Gemini 4 Carbon ranked #1 on AA-Omniscience

    Current1 sourceAA-Omniscience
  6. Artificial AnalysisLISTS PRICE FORGemini 4 Carbon

    Artificial Analysis LISTS_PRICE_FOR Gemini 4 Carbon $4/$20 (50% launch discount)

    Current1 sourceArtificial Analysis
  7. Gemini 4 CarbonCOMPARED WITHGPT-6 Astra

    Gemini 4 Carbon compared with GPT-6 Astra on FrontierSWE (score 61.2% vs 65.5%)

    Current1 sourceGPT-6 Astra
  8. Gemini 4 CarbonRANKED TOP ONAP News

    Gemini 4 Carbon ranked #1 on AutomationBench

    Current1 sourceAP News
  9. Gemini 4 CarbonCOMPARED WITHOpus 5.5

    Gemini 4 Carbon compared with Opus 5.5 on FrontierCode v1.1 (score 51% vs 56%)

    Current1 sourceOpus 5.5
  10. Fairwind ProgramQUOTES PRICE FORGemini 4 Carbon

    Phil Schmid QUOTES_PRICE_FOR Gemini 4 Carbon $2/$10 per million input/output tokens

    Current1 sourceFairwind Program
  11. Vals AICACHED INPUT COST FORGemini 4 Carbon

    ValsAI CACHED_INPUT_COST_FOR Gemini 4 Carbon $0.20 per million tokens

    Current1 sourceVals AI
  12. Fairwind ProgramPROVIDES ACCESS TOGemini 4 Carbon

    Fairwind Program provides access to Gemini 4 Carbon

    Current1 sourceFairwind Program
  13. Gemini 4 CarbonRANKED TOP ONNADA NEWS

    Gemini 4 Carbon ranked #1 on Blueprint-Bench 2

    Current1 sourceNADA NEWS
  14. Vals AISPEC FORGemini 4 Carbon

    ValsAI SPEC_FOR Gemini 4 Carbon 1M-token context window, 1M max output; input modes: text, image, video, speech; output: text

    Current1 sourceVals AI
  15. Gemini 4 CarbonRANKED TOP ONNewsis

    Gemini 4 Carbon ranked #1 on TB-Science

    Current1 sourceNewsis
  16. Gemini 4 CarbonCOMPARED WITHClaude Code

    Gemini 4 Carbon compared with Claude Code on Coding Agent Index (score 59 vs 62)

    Current1 sourceClaude Code
  17. Gemini 4 CarbonRANKED TOP ONHLE-Diamond

    Gemini 4 Carbon ranked #1 on HLE-Diamond

    Current1 sourceHLE-Diamond
  18. Gemini 4 CarbonAVAILABLE INFairwind Program

    Gemini 4 Carbon is currently only available to a select group of cybersecurity defense companies in the Fairwind Program

    Ended 2026-10-291 sourceFairwind Program
  19. GoogleRELEASESGemini 4 Carbon

    Google released Gemini 4 Carbon

    Ended 2026-10-291 sourceGoogle
  20. Gemini 4 CarbonRANKS TOPExploitGym

    Gemini 4 Carbon holds the number one spot on ExploitGym

    Current1 sourceExploitGym
  21. Gemini 4 CarbonIMPROVES OVERGemini 4 Argon

    Gemini 4 Carbon offers significant gains over Argon in coding, agentic tasks, and cybersecurity

    Current1 sourceGemini 4 Argon
  22. Gemini 4 CarbonMULTIPLES REASONING TOKENS THANGemini 4 Argon

    Gemini 4 Carbon generates 2.3 times as many reasoning tokens as Gemini 4 Argon

    Current1 sourceGemini 4 Argon
  23. Gemini 4 CarbonOVERTOPGemini 4 Argon

    Gemini 4 Carbon overtops Gemini 4 Argon in the Text Arena (1,528 vs 1,520 points)

    Current1 sourceGemini 4 Argon
  24. Gemini 4 CarbonMULTIPLES OUTPUT TOKENS THANGemini 4 Argon

    Gemini 4 Carbon generates 64% more output tokens than Gemini 4 Argon

    Current1 sourceGemini 4 Argon
  25. Gemini 4 CarbonTOPS INFinance Index

    Gemini 4 Carbon tops the Finance Index

    Current1 sourceFinance Index
  26. Gemini 4 CarbonSECOND INAgent Arena

    Gemini 4 Carbon is second in the Agent Arena with 1,528 points

    Current1 sourceAgent Arena
  27. Gemini 4 CarbonLOWER TASK COST THANClaude Opus 5

    Gemini 4 Carbon has a median task cost of $1.80, which is 72% lower than Claude Opus 5.5

    Current1 sourceClaude Opus 5
  28. Gemini 4 CarbonLEADS INVals AI

    Gemini 4 Carbon leads in the Vals AI platform, topping both the Finance Index and the Legal Research benchmark

    Current1 sourceVals AI
  29. DeepMindRELEASESGemini 4 Carbon

    Google DeepMind releases Gemini 4 Carbon

    Ended 2026-10-291 sourceDeepMind
  30. Gemini 4 CarbonHIGHER ACCURACY THANClaude Opus 5

    Gemini 4 Carbon has 70.36% accuracy in the Vals Index v2.2, higher than Claude Opus 5.5 70.18% accuracy

    Current1 sourceClaude Opus 5
  31. Gemini 4 CarbonLOWER ACCURACY THANClaude Sonnet 5

    Gemini 4 Carbon has 70.36% accuracy in the Vals Index v2.2, lower than Claude Sonnet 5.5 70.85% accuracy

    Current1 sourceClaude Sonnet 5
  32. Gemini 4 CarbonTOPS INLegal Research benchmark

    Gemini 4 Carbon tops the Legal Research benchmark

  33. Gemini 4 CarbonIS TWICE AS LIKELY TO GENERATE CODE FREE OF VULNERABILITIES ASGemini 4 Argon

    Gemini 4 Carbon is apparently twice as likely to generate code that is free of vulnerabilities as Gemini 4 Argon.

    Current1 sourceGemini 4 Argon
  34. Gemini 4 CarbonIS BETTER THANGemini 4 Argon

    Gemini 4 Carbon is a slightly better version of Gemini 4 Argon.

    Ended 2026-10-291 sourceGemini 4 Argon
  35. GoogleANNOUNCEDGemini 4 Carbon

    Google announced Gemini 4 Carbon.

    Current1 sourceGoogle
  36. Gemini 4 CarbonBEATSClaude Fable 5.1

    Gemini 4 Carbon beats Claude Fable 5.1 in almost every test.

    Current1 sourceClaude Fable 5.1
  37. Gemini 4 CarbonBEATSGPT-6 Astra

    Gemini 4 Carbon beats GPT-6 Astra in almost every test.

    Ended 2026-10-291 sourceGPT-6 Astra
  38. Gemini 4 CarbonGENERATED BYGoogle AI

    This content is generated by Google AI.

    Current1 sourceGoogle AI
  39. Gemini 4 CarbonRANKS INArtificial Analysis Intelligence Index

    Gemini 4 Carbon ranks #1 on the recently updated v5.0 of Artificial Analysis Intelligence Index at 59.

  40. Gemini 4 CarbonAVAILABLE THROUGHFairwind Program

    Gemini 4 Carbon will be available to a subset of cyber defenders and will gradually phase access to additional trusted partners through our Fairwind Program.

    Ended 2026-10-291 sourceFairwind Program

Relations are facts Graphiti extracted from reports, not verbatim quotes. An end date means later reporting showed the relation no longer holds.