BlogLLM Benchmark Scores as Published, Side by Side
GPT-6 Astra vs Gemini 4 Carbon: Benchmark Scores Side by Side
Every test both GPT-6 Astra by OpenAI and Gemini 4 Carbon by Google have a published number for: 18 tests in 11 areas. Each number is as printed, with the page it comes from and the setup it was run under; nothing is ranked or rescored, and two numbers under different setups are two results, not one comparison.
Models GPT-6 Astra, Gemini 4 CarbonData version 2026103001, as of 2026-10-30
Model
What each developer prints on its own model page; a dash means it prints none.
| Fact | GPT-6 Astra | Gemini 4 Carbon |
|---|---|---|
| Released | 2026-09-03 | 2026-10-28 |
| Context window | 1.05M | - |
| Max output | 128K | - |
| Input | Text, image | - |
| Output | Text | - |
| Weights | Closed | - |
Knowledge Work
| Test | GPT-6 Astra | Gemini 4 Carbon |
|---|---|---|
| AutomationBench | 41.4%OpenAI, GPT-6 Astra, checked 2026-10-14; printed on 7 more pages | 53.9%Google, Gemini 4 Carbon, checked 2026-10-29 |
| Vending-Bench 2 (final balance in USD) | $15515Google, Gemini 4 Carbon, checked 2026-10-29 | $16484Google, Gemini 4 Carbon, checked 2026-10-29 |
| Vals Index v2.2 | 65%Google, Gemini 4 Carbon, checked 2026-10-29 | 70.4%Google, Gemini 4 Carbon, checked 2026-10-29 |
Agentic Coding
| Test | GPT-6 Astra | Gemini 4 Carbon |
|---|---|---|
| Terminal-Bench 4.0 | 57.9%OpenAI, GPT-6 Astra, checked 2026-10-14; printed on 7 more pages | 65.2%Google, Gemini 4 Carbon, checked 2026-10-29 |
| FrontierSWE v2 | 65.5%Anthropic, Claude Opus 5.5 System Card, checked 2026-10-25; printed on 3 more pages | 61.2%Google, Gemini 4 Carbon, checked 2026-10-29 |
| SWE-sweep | 9.5%Google, Gemini 4 Carbon, checked 2026-10-29 |
Science and Math
| Test | GPT-6 Astra | Gemini 4 Carbon |
|---|---|---|
| Terminal-Bench Science 0.1 | 64.6%OpenAI, GPT-6 Astra, checked 2026-10-14; printed on 7 more pages | 64.8%Google, Gemini 4 Carbon, checked 2026-10-29 |
| RiemannBench | 72%Google, Gemini 4 Argon, checked 2026-10-29; printed on 1 more page | 80%Google, Gemini 4 Carbon, checked 2026-10-29 |
Long Context
| Test | GPT-6 Astra | Gemini 4 Carbon |
|---|---|---|
| GraphWalks BFS 256k to 1M (F1) | 71.8%Google, Gemini 4 Argon, checked 2026-10-29; printed on 1 more page | 96.2%Google, Gemini 4 Carbon, checked 2026-10-29 |
Computer Use
| Test | GPT-6 Astra | Gemini 4 Carbon |
|---|---|---|
| OSWorld 2.1 offline subset | 80.9%Google, Gemini 4 Carbon, checked 2026-10-29 |
Multimodal Understanding
| Test | GPT-6 Astra | Gemini 4 Carbon |
|---|---|---|
| Chartography (no tools) | 72.5%Google, Gemini 4 Carbon, checked 2026-10-29 |
Cybersecurity
| Test | GPT-6 Astra | Gemini 4 Carbon |
|---|---|---|
| ExploitGym | 42.4%OpenAI, GPT-6 Astra, checked 2026-10-14; printed on 4 more pages | |
| SEC-Bench Pro | 85.4%OpenAI, GPT-6 Astra, checked 2026-10-14; printed on 4 more pages | |
| CWE-bench v1 | 68%Google, Gemini 4 Argon, checked 2026-10-29; printed on 2 more pages | 71%Google, Gemini 4 Carbon, checked 2026-10-29; printed on 1 more page |
Tool Use and Workflows
| Test | GPT-6 Astra | Gemini 4 Carbon |
|---|---|---|
| Agents' Last Exam v1.1 (score) | 73.1%Google, Gemini 4 Carbon, checked 2026-10-29 |
Safety and Alignment
| Test | GPT-6 Astra | Gemini 4 Carbon |
|---|---|---|
| Gray Swan IPI (attack success rate after k attempts)attack success rate after k attempts, lower is better | 8.5%Anthropic, Claude Haiku 5.5 System Card, checked 2026-10-19; printed on 2 more pages | 0%Google, Gemini 4 Carbon: our next frontier model, checked 2026-10-29 |
Engineering and Design
| Test | GPT-6 Astra | Gemini 4 Carbon |
|---|---|---|
| Blueprint-Bench 2 (normalized score) | 49.7%Google, Gemini 4 Carbon, checked 2026-10-29 | 58.6%Google, Gemini 4 Carbon, checked 2026-10-29 |
Preference and Composite Rankings
| Test | GPT-6 Astra | Gemini 4 Carbon |
|---|---|---|
| Artificial Analysis Intelligence Index v5.0 | 57Google, Gemini 4 Carbon, checked 2026-10-29 | 59Google, Gemini 4 Carbon, checked 2026-10-29; printed on 1 more page |
Price
API list prices in US dollars per one million tokens, as printed on the developer's pricing page on the day checked; a dash means the developer lists that price in another unit or not at all. Cost per task is measured by Vals AI.
| Price | GPT-6 Astra | Gemini 4 Carbon |
|---|---|---|
| Input | $10OpenAI, OpenAI API pricing, checked 2026-10-15Context: up to 272,000 tokens; Long context: $20 per 1M tokens above 272,000 tokens | |
| Cached input | $1OpenAI, OpenAI API pricing, checked 2026-10-15Context: up to 272,000 tokens; Long context: $2 per 1M tokens above 272,000 tokens | - |
| Cache write | $12.5OpenAI, OpenAI API pricing, checked 2026-10-15Context: up to 272,000 tokens; Long context: $25 per 1M tokens above 272,000 tokens | - |
| Output | $50OpenAI, OpenAI API pricing, checked 2026-10-15Context: up to 272,000 tokens; Long context: $75 per 1M tokens above 272,000 tokens | |
| Input, batch | $5OpenAI, OpenAI API pricing, checked 2026-10-15Context: up to 272,000 tokens; Long context: $10 per 1M tokens above 272,000 tokens | - |
| Cached input, batch | $0.5OpenAI, OpenAI API pricing, checked 2026-10-15Context: up to 272,000 tokens; Long context: $1 per 1M tokens above 272,000 tokens | - |
| Cache write, batch | $6.25OpenAI, OpenAI API pricing, checked 2026-10-15Context: up to 272,000 tokens; Long context: $12.5 per 1M tokens above 272,000 tokens | - |
| Output, batch | $25OpenAI, OpenAI API pricing, checked 2026-10-15Context: up to 272,000 tokens; Long context: $37.5 per 1M tokens above 272,000 tokens | - |
| Input, flex | $5OpenAI, OpenAI API pricing, checked 2026-10-15Context: up to 272,000 tokens; Long context: $10 per 1M tokens above 272,000 tokens | - |
| Cached input, flex | $0.5OpenAI, OpenAI API pricing, checked 2026-10-15Context: up to 272,000 tokens; Long context: $1 per 1M tokens above 272,000 tokens | - |
| Cache write, flex | $6.25OpenAI, OpenAI API pricing, checked 2026-10-15Context: up to 272,000 tokens; Long context: $12.5 per 1M tokens above 272,000 tokens | - |
| Output, flex | $25OpenAI, OpenAI API pricing, checked 2026-10-15Context: up to 272,000 tokens; Long context: $37.5 per 1M tokens above 272,000 tokens | - |
| Input, fast | $20OpenAI, OpenAI API pricing, checked 2026-10-15Context: up to 272,000 tokens; Long context: $40 per 1M tokens above 272,000 tokens | - |
| Cached input, fast | $2OpenAI, OpenAI API pricing, checked 2026-10-15Context: up to 272,000 tokens; Long context: $4 per 1M tokens above 272,000 tokens | - |
| Cache write, fast | $25OpenAI, OpenAI API pricing, checked 2026-10-15Context: up to 272,000 tokens; Long context: $50 per 1M tokens above 272,000 tokens | - |
| Output, fast | $100OpenAI, OpenAI API pricing, checked 2026-10-15Context: up to 272,000 tokens; Long context: $150 per 1M tokens above 272,000 tokens | - |
| Input, ultrafast | $60OpenAI, OpenAI API pricing, checked 2026-10-15Context: up to 272,000 tokens; Long context: $120 per 1M tokens above 272,000 tokens | - |
| Cached input, ultrafast | $6OpenAI, OpenAI API pricing, checked 2026-10-15Context: up to 272,000 tokens; Long context: $12 per 1M tokens above 272,000 tokens | - |
| Cache write, ultrafast | $75OpenAI, OpenAI API pricing, checked 2026-10-15Context: up to 272,000 tokens; Long context: $150 per 1M tokens above 272,000 tokens | - |
| Output, ultrafast | $300OpenAI, OpenAI API pricing, checked 2026-10-15Context: up to 272,000 tokens; Long context: $450 per 1M tokens above 272,000 tokens | - |
| Cost per taskVals AI, USD, lower is better | - |
GPT-6 Astra has 184 more and Gemini 4 Carbon has 3 more: tests only one of them has a number for, listed on its model page.
More side by side
Every number here is read from the same catalog file as the benchmark table, data version 2026103001. A pair has a page when both models are their developers' flagships and both have a number for at least 10 of the same tests. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.