Public benchmarkExternal source reviewReviewed by DanPublished Source checked Updated
Gemini 4 Carbon: benchmark source and usage profile
The report
Verdict
Published benchmark evidence
Carbon's published numbers span at least two evaluation operators and distinct harnesses. The 65.2% terminal result should not be compared directly with another model's non-AA Terminal-Bench score.
On the tier list: The November 17 tier is Dan’s own ranking; public benchmark numbers are a separate evidence channel, not an automatic tier adjustment.
What the benchmarks report
- Terminal-Bench 4.0 (AA): 65.2%. Carbon; Artificial Analysis evaluation. Reported by Google DeepMind vendor model card.
- FrontierSWE v2: 61.2%. Carbon; agentic coding. Reported by Google DeepMind vendor model card.
- FrontierCode 2.0 — Cognition: 42.6%. Carbon; high; Antigravity CLI harness. Reported by Cognition benchmark operator.
Original sources
Source 1 — Terminal-Bench 4.0 (AA), FrontierSWE v2.
Source 2 — FrontierCode 2.0 — Cognition.
What this evidence tells a reader
The 65.2% Terminal-Bench 4.0 value explicitly names Artificial Analysis. That qualifier is not decoration: a different harness or evaluation track can change the result even when the suite version looks the same. For example, Argon's model card reports a different terminal configuration, so these two visible figures do not by themselves establish a Carbon-versus-Argon winner.
FrontierSWE v2 at 61.2% covers agentic software engineering; Cognition's FrontierCode 2.0 at 42.6% uses a high-effort Antigravity CLI setup. Rather than treat the numbers as inconsistent, read them as evidence about different task pools and implementations. Neither tells a developer how Carbon will handle an existing project's secrets, review process or tool permissions.
A practical trial should use the exact Carbon checkpoint in the intended harness. A provider alias or free-access gateway may not expose the same conditions as the published model-card result.
How to use these results
- Keep the AA label attached to the terminal result wherever it is quoted.
- Before a coding purchase or migration, verify checkpoint, effort, harness and usage terms, then run one reviewed project issue.
- Prefer directly matched evaluation methods when comparing Carbon with Argon or another Gemini model.
Limits of this report
- Google's model-card figures and Cognition's Antigravity run are not interchangeable evaluation conditions.
- This page does not verify a local Carbon checkpoint, free-provider behavior, token cost, or tool reliability.
Follow the lab
New test reports and tier-list changes get announced in my Discord's #news channel. The full videos are on YouTube.
Receipts
DanManREAL · AI lab
Test receipt
Gemini 4 Carbon
- Terminal-Bench 4.0 (AA)
- 65.2%Google DeepMind vendor model card (model publisher). Carbon; Artificial Analysis evaluation. Not independently rerun by DanManREAL.
- FrontierSWE v2
- 61.2%Google DeepMind vendor model card (model publisher). Carbon; agentic coding. Not independently rerun by DanManREAL.
- FrontierCode 2.0 — Cognition
- 42.6%Cognition benchmark operator (independent operator or secondary publication). Carbon; high; Antigravity CLI harness. Not independently rerun by DanManREAL.
- Source review
- November 17, 2026Date these source claims were reviewed; not a claim about when the published benchmark was executed.
Verdict
Published benchmark evidence
Tier list The November 17 tier is Dan’s own ranking; public benchmark numbers are a separate evidence channel, not an automatic tier adjustment.
danmanreal.com/tests/gemini-4-carbon-public-evidence
Models covered by external benchmarks
On the tier list's bench
- Gemini 4 Carbon · 1 receipt on file
Related
Independent testing. Sponsorships are always disclosed, and verdicts are never for sale.
Disagree? Bring your receiptsAll reports
- One inbox for my AI coding agents: triage took 34% less time, resuming work did not improveHead-to-headNov 20
- Mythos 5.2: research and security benchmark source reviewPublic benchmarkNov 17
- Fable 5.2: agentic coding benchmark source reviewPublic benchmarkNov 17
- Claude Opus 5.5: public coding benchmark receiptsPublic benchmarkNov 17
- Astra 6.2: new model and public task-evaluation evidencePublic benchmarkNov 17
- Sol 6.2: new model and public coding benchmark evidencePublic benchmarkNov 17
- Gemini 4 Carbon: benchmark source and usage profilePublic benchmarkNov 17
- Qwen 4: Qwen4-Max launch and benchmark sourcesPublic benchmarkNov 17
- Kimi K3.1 Pro: public coding benchmark receiptsPublic benchmarkNov 17
- GLM-5.3: public coding benchmark receiptsPublic benchmarkNov 17
- Gemini 4 Argon: public coding benchmark receiptsPublic benchmarkNov 17
- Claude Sonnet 5.5: public coding benchmark receiptsPublic benchmarkNov 17
- DeepSeek V4.1 Flash: public coding benchmark receiptsPublic benchmarkNov 17
- Muse Wonder Full 1.0: public coding benchmark receiptsPublic benchmarkNov 17
- Grok 4.7: separate release and public coding resultsPublic benchmarkNov 17
- ai-design-check on two of my FABLE screens: 10 warnings, 4 of them false alarmsField testNov 2
- Vercel Agent Desktop used a sample Trip app in 6 actions without moving the mouseField testNov 1
- Cerberus Guard on sample data: one crawler name, two verdicts, and a block earned by behaviorField testOct 31
- GPT-6.1 Terra vs Opus 5.5: three real projects, and the designs I picked blindHead-to-headOct 29
- Steps → Reduced: opening a terminal agent went from 98 seconds to 11AuditOct 28
- One ChatGPT chat directed Opus and Codex at once through DMR_REMOTEField testOct 28
- Claude Duet ran my channel for a week and caught 269 fake AI crawlersField testOct 27
- Kimi K3.1 Pro built my animation family, after a quota stop and 21 minutes of failed requestsField testOct 24
- ChatGPT + Codex built 33 new DMR_REMOTE toolsField testOct 23
- Gemini 4 Argon reviewed a finished episode: 14 clips, one flag, one inconsistencyField testOct 21
- Qwen 3.8 Flash Next on my RTX 5080: 81 tokens a second with 202K tokens readAuditOct 21
- Does an extra AI review step pay off? A small DeepSeek trial said noHead-to-headOct 19
- My real DeepSeek bill: 12.31 billion tokens for $122.28Cost receiptOct 17
- Gemini 4 Argon kept hanging in Gemini CLI. The cause was a 60-second timeoutIncidentOct 14
- I asked 10 AI coding agents for their biggest weaknessSelf-reportOct 14
- Jarvis 27B on my own PC: 30 tokens a second, and it still recalls at 251KAuditOct 10
- My agents' first game benchmark was wrong. Here's how it was caughtIncidentOct 10
- Pixal3D + Blender turned my 2D art into animated 3D charactersField testSep 22
- DeepSeek Flash answered 200 OK and sent nothing for two hoursIncidentSep 14
- Grok, Codex and DeepSeek played WorldOS for Rank 1. Grok won with 98Head-to-headAug 20
- DeepSeek V4 Flash audited 18 of my real coding sessions. Two said 'done' too earlyAuditAug 4