Gemini 4 Argon vs GPT-6 Astra
Full profiles: Gemini 4 Argon (Google) and GPT-6 Astra (OpenAI): every benchmark, price and rank.
Updated · Data from 10 sources
As of 21 Oct 2026, Gemini 4 Argon (high) and GPT-6 Astra (max) are neck and neck on quality (71.0 vs 70.6 on the BenchLeader Index, inside the ±4.5 margin of error), and Gemini 4 Argon (high) costs 5.0× less, which makes it the better buy for most workloads. Gemini 4 Argon (high)'s current price runs until 30 Nov 2026; from 1 Dec 2026 Gemini 4 Argon (high) costs 2.5× less ($8.00 vs $20.00 per 1M blended); on use case, pick Gemini 4 Argon for agents & tools (72 vs 65), GPT-6 Astra for knowledge (79 vs 77).
Gemini 4 Argon benchmarks 
GPT-6 Astra benchmarks Which to pick for…
The pick's numbers come first. Category scores on the BenchLeader scale (50 is the average model, 15 points one standard deviation). High confidence means a lead of half a standard deviation or more; under one point is called either.
| Use case | Pick | Why | Confidence |
|---|---|---|---|
| Coding | GPT-6 Astra | 74 vs 72 category score; Arena WebDev 1774 vs 1678 | Medium |
| Agents & tool use | Gemini 4 Argon | 72 vs 65 category score; Arena Agent 12.1 vs 15.1 | Medium |
| Maths | GPT-6 Astra | 72 vs 71 category score; FrontierMath Tiers 1–3 93.7% vs 86.0% | Medium |
| Reasoning | GPT-6 Astra | 76 vs 75 category score; ARC-AGI-2 95.0% vs 91.3% | Medium |
| Knowledge | GPT-6 Astra | 79 vs 77 category score; AA-Omniscience 43.4 vs 42.4 | Medium |
| Long documents | Either | 65 vs 65 category score; AA-LCR 79.7% vs 80.7% | Either |
| Chat quality (human preference) | Gemini 4 Argon | 73 vs 67 on Arena votes | Medium |
| Lowest cost | Gemini 4 Argon | $4.00 vs $20.00 per 1M blended; agentic coding session $0.160 vs $0.800; from 1 Dec 2026: $8.00 vs $20.00 | – |
| Fastest first answer | Gemini 4 Argon | 48.5 s (Gemini 4 Argon (high)) vs 328.5 s (GPT-6 Astra (max)) to the first answer token | – |
| High volume on a budget | Neither flagship | A smaller model is usually the better fit: Cheapest good LLMs, Fastest LLMs | – |
Key differences
- Gemini 4 Argon (high) and GPT-6 Astra (max) are neck and neck on quality (71.0 vs 70.6, within the ±4.5 margin).
- Gemini 4 Argon (high) is stronger in agents & tools, human preference.
- GPT-6 Astra (max) is stronger in coding, composite, knowledge, long context, maths, reasoning, multimodal.
- Gemini 4 Argon (high) is 5.0× cheaper ($4.00 vs $20.00 per 1M blended) until 30 Nov 2026, then $8.00 vs $20.00 from 1 Dec 2026.
At a glance
| Spec | Gemini 4 Argon | GPT-6 Astra |
|---|---|---|
| Developer | OpenAI | |
| Released | 30 Sept 2026 | 2 Sept 2026 |
| Generally available | 13 Oct 2026 | 2 Sept 2026 |
| Context window (tokens) | 1,048,576 | 1,050,000 |
| Max output (tokens) | 65,536 (up to 1,048,576 with long decode) | – |
| Input types | text, image, video, audio, PDF | – |
| Reasoning effort settings | low, medium, high | – |
| Compared at | Gemini 4 Argon (high) | GPT-6 Astra (max) |
| Open weights | No | No |
| Price in / out / cached input per 1M | $2.00 / $10.00 / $0.100 | $10.00 / $50.00 / $1.00 |
| Announced price change | $4.00 / $20.00 / $0.200 from 1 Dec 2026 | None announced |
| API model ID | gemini-4-argon | – |
| Where to use it | Gemini API (AI Studio), Gemini Enterprise Agent Platform, Gemini app | – |
Limits, availability and announced prices from the vendors' own pages: ai.google.dev, ai.google.dev pricing (checked 20 Oct 2026). Everything else is from the day's data.
Category by category
Category scores on the BenchLeader scale: 50 is the average evaluated model, 15 points is one standard deviation. A gap under one point is called level.
| Category | Gemini 4 Argon (high) | GPT-6 Astra (max) | Leader |
|---|---|---|---|
| Reasoning | 74.9 | 76.3 | GPT-6 Astra |
| Coding | 72.2 | 74.0 | GPT-6 Astra |
| Agents & tools | 72.0 | 64.8 | Gemini 4 Argon |
| Maths | 70.5 | 72.0 | GPT-6 Astra |
| Knowledge | 76.5 | 79.4 | GPT-6 Astra |
| Human preference | 72.9 | 67.2 | Gemini 4 Argon |
| Multimodal | – | 66.1 | not comparable |
| Long context | 64.9 | 65.4 | level |
| Composite | 88.5 | 88.6 | level |
Where sources disagree
Across the 61 benchmarks both have, Gemini 4 Argon leads 25, GPT-6 Astra leads 34 and 2 are tied. Independent sources split: Gemini 4 Argon leads on Vals Index and Arena Text; GPT-6 Astra leads on AA Intelligence Index v4.3.2.
| Overall index | Gemini 4 Argon (high) | GPT-6 Astra (max) | Leader |
|---|---|---|---|
| BenchLeader Index | 71.0 | 70.6 | neck and neck |
| AA Intelligence Index v4.3.2 | 52.6 | 52.7 | GPT-6 Astra |
| Vals Index | 69.2 | 65.0 | Gemini 4 Argon |
| Arena Text | 1524 | 1477 | Gemini 4 Argon |
Each publisher runs its own tasks, harnesses and effort settings, so their overall verdicts can differ; BenchLeader reconciles them on one scale. † measured at another effort setting of the same model (the publisher did not test the one compared here).
Price, speed and latency
List prices blended at three input tokens per output token. Per-task prices use the token counts on the methodology page; suite cost is what Artificial Analysis paid to run its whole Intelligence Index.
| Measure | Gemini 4 Argon (high) | GPT-6 Astra (max) |
|---|---|---|
| Blended price per 1M tokens, until 30 Nov 2026 | $4.00 | $20.00 |
| Input / output per 1M | $2.00 / $10.00 | $10.00 / $50.00 |
| Blended price per 1M from 1 Dec 2026 | $8.00 | $20.00 |
| Input / output per 1M from 1 Dec 2026 | $4.00 / $20.00 | $10.00 / $50.00 |
| Cached input per 1M | $0.100 | $1.00 |
| Price per task: chat reply | $0.0038 | $0.019 |
| Price per task: summarize a 30-page report | $0.030 | $0.150 |
| Price per task: code edit | $0.027 | $0.135 |
| Price per task: agentic coding session | $0.160 | $0.800 |
| Price per task: structured extraction | $0.0060 | $0.030 |
| Cost to run a full benchmark suite | $1.99 | $3.26 |
| Output speed | 83 tok/s | 64 tok/s |
| Time to first answer | 48.5 s | 328.5 s |
| Time to a complete answer | 54.6 s | 336.3 s |
Per token is not per task: on Artificial Analysis' full suite Gemini 4 Argon (high) cost $1.99 and GPT-6 Astra (max) $3.26, because Gemini 4 Argon (high) used about 3.1× as many tokens for the same work, which a per-token price hides.
Each model is compared at its best-ranked reasoning-effort setting. Gemini 4 Argon (high) scores 71.0 and takes 49 s to its first answer token; Gemini 4 Argon scores 63.3 in 4 s. GPT-6 Astra (max) scores 70.6 and takes 328 s to its first answer token; GPT-6 Astra (high) scores 69.7 in 64 s, GPT-6 Astra (xhigh) scores 69.0 in 159 s. Higher effort means more thinking before the answer: better scores, longer waits and more tokens billed. How reasoning effort changes scores and cost.
Every benchmark
125 benchmarks, grouped by category; each name links to the board and its leaders. † measured at another effort setting of the same model, shown for reference and not counted as a win.
| Metric | Publisher | Gemini 4 Argon (high) | GPT-6 Astra (max) |
|---|---|---|---|
| BenchLeader Index | BenchLeader | 71.0 | 70.6 |
| Agents & tools score | BenchLeader | 72.0 | 64.8 |
| Coding score | BenchLeader | 72.2 | 74.0 |
| Composite score | BenchLeader | 88.5 | 88.6 |
| Human preference score | BenchLeader | 72.9 | 67.2 |
| Knowledge score | BenchLeader | 76.5 | 79.4 |
| Long context score | BenchLeader | 64.9 | 65.4 |
| Maths score | BenchLeader | 70.5 | 72.0 |
| Reasoning score | BenchLeader | 74.9 | 76.3 |
| Multimodal score | BenchLeader | – | 66.1 |
| Blended price $/M | BenchLeader | $4.00 | $20.00 |
| Output speed | BenchLeader | 83 tok/s | 64 tok/s |
| Time to first answer | BenchLeader | 48.5 s | 328.5 s |
| Context window | BenchLeader | 1,048,576 tokens | 1,050,000 tokens |
| Reasoning | |||
| GPQA Diamond | Epoch AI Benchmarking Hub | 96.4% | 95.8% |
| Humanity's Last Exam | Scale AI / CAIS | – | 54.8%† |
| Arena Hard Prompts | Arena (LMArena) | 1549 | 1502 |
| LiveBench 2.0 Reasoning | LiveBench | 95.1%† | 92.7%† |
| LiveBench 1.0 Reasoning | LiveBench | – | 92.7% |
| GPQA Diamond (AA) | Artificial Analysis | – | 96.1% |
| Humanity's Last Exam (AA) | Artificial Analysis | 57.1% | 54.7% |
| ARC-AGI-1 | ARC Prize | 98.5% | 97.5% |
| ARC-AGI-2 | ARC Prize | 91.3% | 95.0% |
| ARC-AGI-3 | ARC Prize | 97.3% | 98.6% |
| CritPt | Artificial Analysis | 27.1% | 31.7% |
| MysteryMechanism | Vals AI | 45.5% | 53.1% |
| Chess Puzzles | Epoch AI Benchmarking Hub | 59.0% | 74.0% |
| Mystery Game Puzzles | Epoch AI Benchmarking Hub | 64.0% | 84.0% |
| LMCA | Epoch AI Benchmarking Hub | 63.7% | 64.1%† |
| DTBench | Epoch AI Benchmarking Hub | 97.3% | 97.1%† |
| BTF-3 | Epoch AI Benchmarking Hub | – | 13.5%† |
| Coding | |||
| SciCode | SciCode | 61.8% | 56.5% |
| WeirdML | WeirdML | – | 93.3% |
| FrontierCode | Cognition | 47.5%† | 53.3% |
| GSO-Bench | GSO-Bench | – | 79.4%† |
| Arena Coding | Arena (LMArena) | 1559 | 1543 |
| Arena WebDev | Arena (LMArena) | 1678 | 1774 |
| LiveBench 2.0 Coding | LiveBench | 89.7%† | 93.7%† |
| LiveBench 1.0 Coding | LiveBench | – | 80.4% |
Show all 125 benchmarks
| Metric | Publisher | Gemini 4 Argon (high) | GPT-6 Astra (max) |
|---|---|---|---|
| Coding | |||
| IOI | Vals AI | 100.0% | 100.0% |
| Code Migration | Vals AI | 68.2% | 67.7% |
| ProgramBench | Vals AI | 2.5% | 5.5% |
| Vibe Code Bench 1-100 | Vals AI | 21.2% | 27.6% |
| Vibe Code Bench v1.1 | Vals AI | 91.9% | 89.6% |
| SWE Atlas: Codebase QnA | Scale AI SEAL | – | 59.1%† |
| SWE Atlas: Refactoring | Scale AI SEAL | – | 59.0%† |
| SWE Atlas: Test Writing | Scale AI SEAL | – | 50.7%† |
| MirrorCode | Epoch AI Benchmarking Hub | – | 46.7%† |
| DeepSWE v1.1 | Epoch AI Benchmarking Hub | – | 73.2% |
| CursorBench | Cursor | 52.9% | – |
| ALE-Bench | Epoch AI Benchmarking Hub | 2100.1 | 2951.3 |
| FrontierSWE | Epoch AI Benchmarking Hub | 55.0% | 65.5% |
| Terminal-Bench 4.0 (AA) | Artificial Analysis | 57.1% | 59.1% |
| Terminal-Bench 2.1 (AA) | Artificial Analysis | – | 88.4% |
| Startrise Frontend | Startrise | – | 79.9%† |
| FrontierCode 1.1 (AA) | Artificial Analysis | 43.5% | 47.0% |
| Agents & tools | |||
| Terminal-Bench | Terminal-Bench | – | 58.2% |
| Remote Labor Index | Scale AI / CAIS | – | 20.8%† |
| APEX-Agents | Mercor | 82.4%† | 64.7%† |
| Arena Agent | Arena (LMArena) | 12.1 | 15.1 |
| LiveBench Long Agentic | LiveBench | 69.9%† | 82.4%† |
| LiveBench Browser | LiveBench | 92.0%† | 84.0%† |
| LiveBench Agentic Coding | LiveBench | – | 57.3% |
| Terminal-Bench 2.1 (Vals) | Vals AI | – | 87.3% |
| HiL-Bench | Scale AI SEAL | 62.0%† | 65.3%† |
| GDPval-AA v2.1 | Artificial Analysis | 56.2% | 53.6% |
| τ³-Banking (AA) | Artificial Analysis | – | 41.4% |
| ITBench SRE (AA) | Artificial Analysis | – | 48.6% |
| Analyst Agent (AA) | Artificial Analysis | 48.8% | 51.3% |
| CUA-bench | Vals AI | 4.8% | 19.2% |
| CyberBench | Vals AI | 77.9% | 41.1% |
| Finance Agent v2 | Vals AI | 65.4% | 53.5% |
| Harvey's Legal Agent Benchmark | Vals AI | 27.1% | 7.1% |
| Legal Research Bench | Vals AI | 54.8% | 39.4% |
| SREBench | Vals AI | 44.3% | 56.9% |
| Tax Agent Bench | Vals AI | 44.6% | 20.7% |
| Terminal-Bench 4.0 (Vals) | Vals AI | 57.6% | 59.6% |
| Terminal-Bench Science | Vals AI | 44.3% | 62.9% |
| Time Horizon Index: KSP | Vals AI | 52.5% | 90.5% |
| Arena Search | Arena (LMArena) | – | 1246 |
| BALROG | BALROG | – | 68.3% |
| Vending-Bench 2 | Andon Labs | 13718.2† | 15514.7† |
| AutomationBench | Artificial Analysis | 77.5% | 68.5% |
| GDP.pdf (AA) | Artificial Analysis | 21.8% | 31.0% |
| Harvey LAB-AA v1.1 | Harvey | 3.9% | 8.6% |
| AA-Briefcase v1.1 | Artificial Analysis | 1486 | 1569 |
| Terminal-Bench-Science 0.1 (AA) | Artificial Analysis | 52.9% | 63.3% |
| CWE-Bench (AA) | Artificial Analysis | – | 63.3% |
| DeepSec-Bench (AA) | Artificial Analysis | – | 36.9% |
| CyberGym end-to-end (AA) | Artificial Analysis | – | 0.0% |
| FACTS Search | 84.0%† | 89.9%† | |
| Maths | |||
| FrontierMath Tiers 1–3 | Epoch AI Benchmarking Hub | 86.0% | 93.7% |
| FrontierMath Tier 4 | Epoch AI Benchmarking Hub | 78.0% | 97.6% |
| OTIS Mock AIME | Epoch AI Benchmarking Hub | 100.0% | 100.0% |
| ProofBench | Vals AI | 99.0%† | 99.0%† |
| LiveBench Mathematics | LiveBench | – | 96.8% |
| Arena Maths | Arena (LMArena) | 1529 | 1486 |
| Knowledge | |||
| SimpleQA Verified | Epoch AI Benchmarking Hub | 76.0% | 75.6% |
| LiveBench 2.0 Data Analysis | LiveBench | 85.2%† | 83.0%† |
| LiveBench Research | LiveBench | 83.1%† | 85.2%† |
| LiveBench 1.0 Data Analysis | LiveBench | – | 83.0% |
| AA-Omniscience | Artificial Analysis | 42.4 | 43.4 |
| LegalBench | Vals AI | 88.3% | – |
| PRBench Finance | Scale AI SEAL | 49.9%† | 47.5%† |
| PRBench Legal | Scale AI SEAL | 49.4%† | 48.4%† |
| BioMysteryBench | Vals AI | 76.3% | 79.3% |
| Excel Modeling Benchmark | Vals AI | 75.2% | 71.7% |
| MedCode | Vals AI | 58.8% | 48.5% |
| MedScribe | Vals AI | 87.4% | 87.9% |
| Public Benefits Bench | Vals AI | 69.8% | – |
| DrugDiscoveryBench | Scale AI SEAL | – | 68.7% |
| EBR-bench | Epoch AI Benchmarking Hub | 64.8% | 76.2% |
| GDP.pdf | Epoch AI Benchmarking Hub | 28.4% | 34.2% |
| AA-Omniscience: accuracy | Artificial Analysis | 49.9% | 62.6% |
| AA-Omniscience: non-hallucination | Artificial Analysis | 84.9% | 48.7% |
| Instruction following | |||
| LiveBench Language | LiveBench | – | 89.4% |
| LiveBench Instruction Following | LiveBench | – | 75.6% |
| Arena Instruction Following | Arena (LMArena) | 1527 | 1473 |
| Human preference | |||
| Arena Text | Arena (LMArena) | 1524 | 1477 |
| Arena Creative Writing | Arena (LMArena) | 1516 | 1450 |
| Arena Multi-turn | Arena (LMArena) | 1552 | 1486 |
| Arena Longer Queries | Arena (LMArena) | 1543 | 1489 |
| Multimodal | |||
| Arena Vision | Arena (LMArena) | – | 1280 |
| LiveBench Design | LiveBench | 73.7%† | 82.0%† |
| MMMU-Pro | Artificial Analysis | – | 86.9% |
| SAGE | Vals AI | 53.6% | 46.4% |
| AudioMC | Scale AI SEAL | 58.4% | – |
| AudioMC (text output) | Scale AI SEAL | 58.4% | – |
| Blueprint-Bench 2 | Epoch AI Benchmarking Hub | 54.5%† | 49.7%† |
| AA-3D | Artificial Analysis | 1154 | 1166 |
| Long context | |||
| AA-LCR | Artificial Analysis | 79.7% | 80.7% |
| Arena Document | Arena (LMArena) | – | 1473 |
| MLCR | Artificial Analysis | – | 35.0% |
| Composite | |||
| Epoch Capabilities Index | Epoch AI Benchmarking Hub | 166.4† | 168.8† |
| LiveBench 2.0 | LiveBench | 84.1%† | 86.1%† |
| LiveBench 1.0 | LiveBench | – | 82.2% |
| AA Intelligence Index v4.3.2 | Artificial Analysis | 52.6 | 52.7 |
| Vals Index | Vals AI | 69.2 | 65.0 |
| AA Cyber Index | Artificial Analysis | – | 33.4 |
Data as of 2026-10-21. Best configuration of each model; every score links to its source on the model pages.
What changed
- Gemini 4 Argon: list price moves to $4.00 / $20.00 per 1M input / output tokens (announced).
- Gemini 4 Argon (high): 1 new result (Harvey LAB-AA v1.1); 2 score revisions.
- GPT-6 Astra (max): 4 score revisions.
- GPT-6 Astra (max): BenchLeader Index 69.8 → 70.6.
- Gemini 4 Argon (high): 3 new results (LMCA, DTBench, ALE-Bench).
- Gemini 4 Argon (high): 5 new results (Analyst Agent (AA), FrontierCode 1.1 (AA), Terminal-Bench-Science 0.1 (AA)…); 1 score revision.
- Gemini 4 Argon (high): 6 new results (EBR-bench, Time Horizon Index: KSP, Vibe Code Bench 1-100…); 1 score revision.
- GPT-6 Astra (max): 1 score revision.
- Gemini 4 Argon (high): BenchLeader Index 70.5 → 71.1.
- Gemini 4 Argon (high): 10 new results (FrontierMath Tiers 1–3, FrontierMath Tier 4, OTIS Mock AIME…); 1 score revision.
Related comparisons
- Gemini 4 Argon vs Claude Opus 5.5
- Gemini 4 Argon vs Muse Wonder Full
- Gemini 4 Argon vs Claude Fable 5.1
- GPT-6 Astra vs Claude Opus 5.5
- GPT-6 Astra vs Muse Wonder Full
- GPT-6 Astra vs Claude Fable 5.1
- Gemini 3.8 Flash vs GPT-6 Astra
- Gemini 4 Argon vs GPT-5.6 Sol
Rankings: Best LLM overall · Best LLM for coding · Best LLM for agents · Best cheap LLMs
Gemini 4 Argon vs GPT-6 Astra: questions
- Is Gemini 4 Argon better than GPT-6 Astra?
- Gemini 4 Argon (high) and GPT-6 Astra (max) are neck and neck on quality (71.0 vs 70.6, within the ±4.5 margin). The BenchLeader Index combines every independent quality benchmark; the two are within the margin of error as of 2026-10-21, so check the category scores for your use.
- Which is cheaper, Gemini 4 Argon or GPT-6 Astra, and will that change?
- Gemini 4 Argon is cheaper today: $4.00 against $20.00 per million tokens, blended at three input tokens per output token. Yes, that changes: Gemini 4 Argon's list price changes on 1 Dec 2026, and from then Gemini 4 Argon is cheaper ($8.00 vs $20.00 blended).
- Is Gemini 4 Argon better than GPT-6 Astra for coding?
- GPT-6 Astra scores higher in coding (74 vs 72 on the category index, where 50 is average). On Arena WebDev, the top shared coding benchmark, it is 1678 for Gemini 4 Argon against 1774 for GPT-6 Astra.
- Is Gemini 4 Argon better than GPT-6 Astra for agents and tool use?
- Gemini 4 Argon scores higher in agentic tasks (72 vs 65 on the category index, where 50 is average). On Arena Agent, the top shared agents & tools benchmark, it is 12.1 for Gemini 4 Argon against 15.1 for GPT-6 Astra.
- Which is better for long documents?
- They are level on long-context benchmarks (65 vs 65). Both take about 1.1M tokens: Gemini 4 Argon 1,048,576, GPT-6 Astra 1,050,000.
- Which has lower latency?
- Gemini 4 Argon (high) answers first: 48.5 s against 328.5 s to the first answer token, as Artificial Analysis measures it. Reasoning effort drives most of this: a higher setting thinks longer before it answers.
- Which is faster, Gemini 4 Argon or GPT-6 Astra?
- Gemini 4 Argon streams faster: 83 against 64 output tokens per second.
- Do independent benchmarks agree on Gemini 4 Argon vs GPT-6 Astra?
- Across the 61 benchmarks both have, Gemini 4 Argon leads 25, GPT-6 Astra leads 34 and 2 are tied. Independent sources split: Gemini 4 Argon leads on Vals Index and Arena Text; GPT-6 Astra leads on AA Intelligence Index v4.3.2.
- When was Gemini 4 Argon released?
- Gemini 4 Argon became generally available on 13 Oct 2026, after first appearing on 30 Sept 2026.