Public benchmarkExternal source reviewReviewed by DanPublished Source checked Updated
Gemini 4 Argon: public coding benchmark receipts
The report
Verdict
Published benchmark evidence
Argon's published DeepSWE, FrontierSWE and terminal scores describe different work; the 57.4% Terminal-Bench figure is not a matched run against Carbon's AA track.
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
- DeepSWE v1.1: 77.9%. Argon; agentic coding. Reported by Google DeepMind vendor model card.
- Terminal-Bench 4.0: 57.4%. Argon; model-card methodology. Reported by Google DeepMind vendor model card.
- FrontierSWE v2: 55.0%. Argon; agentic coding. Reported by Google DeepMind vendor model card.
Original sources
Source 1 — DeepSWE v1.1, Terminal-Bench 4.0, FrontierSWE v2.
What this evidence tells a reader
Argon's model card lists 77.9% for DeepSWE v1.1 and 55.0% for FrontierSWE v2. Both address software engineering but use different test constructions, so the 22.9-point spread is not a measured penalty for longer or harder tasks. The 57.4% Terminal-Bench 4.0 result covers terminal-agent work under the model card's stated methodology.
Carbon's model card reports 65.2% on a Terminal-Bench 4.0 Artificial Analysis track. It would be misleading to proclaim Carbon better just by comparing that value with Argon's 57.4%, because the report does not establish matched harnesses. The proper comparison would hold suite, effort, tooling and score aggregation steady.
Separate DanManREAL Argon field reports document a CLI timeout and a video-review use case. Those are more directly relevant to command reliability or creative review than the model card's coding percentages, but neither field report reruns the published suites.
How to use these results
- For terminal use, verify the exact Argon endpoint and reproduce a real command with timeout handling.
- For code changes, compare accepted patches against the same task under matched harness settings.
- Read the linked firsthand failure and video-review reports for their own narrow findings, not as confirmation of vendor scores.
Limits of this report
- The Argon and Carbon terminal results use differently labeled methods; no direct Carbon-versus-Argon rank is justified.
- The model card cannot establish behavior in DanManREAL's CLI, including timeout handling and recovery.
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 Argon
- DeepSWE v1.1
- 77.9%Google DeepMind vendor model card (model publisher). Argon; agentic coding. Not independently rerun by DanManREAL.
- Terminal-Bench 4.0
- 57.4%Google DeepMind vendor model card (model publisher). Argon; model-card methodology. Not independently rerun by DanManREAL.
- FrontierSWE v2
- 55.0%Google DeepMind vendor model card (model publisher). Argon; agentic coding. 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-argon-public-benchmarks
Models covered by external benchmarks
On the tier list's bench
- Gemini 4 Argon · 4 receipts on file
Related
- AI Coding Agents Tier List
- Gemini 4 Argon kept hanging in Gemini CLI. The cause was a 60-second timeout
- Gemini 4 Argon reviewed a finished episode: 14 clips, one flag, one inconsistency
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