Gemini 4 Carbon vs GPT-6.1 Sol: OpenAI Needs Code Red Again
Google has officially announced Gemini 4 Carbon, its second flagship model barely two weeks after Gemini 4 Argon became available to everyone. Carbon is a significant jump from Argon, which was already the best at some benchmarks and beat Claude Opus 5.5 in several of them. Here is the detailed comparison between the new Gemini 4 Carbon and OpenAI's GPT-6.1 Sol.
Benchmarks
| Benchmark | Gemini 4 Carbon (High) | GPT-6.1 Sol (Max) | Difference |
|---|---|---|---|
| Intelligence Index v5.0 | 59 | 55 | Carbon +4 |
| AA-Briefcase v1.1 | 1609 Elo | 1555 Elo | Carbon +54 Elo |
| AutomationBench-AA | 78% | 65% | Carbon +13 points |
| Terminal-Bench 4.0 | 65% | 56% | Carbon +9 points |
| FrontierCode 2.0 | 41% | 41% | Tie |
| AA-Omniscience v1.1 | 54 | 42 | Carbon +12 |
| GDP.pdf | 28% | 31% | Sol +3 points |
| HLE-Diamond | 69% | 59% | Carbon +10 points |
| CritPt v2 (Hard Subset) | 63% | 69% | Sol +6 points |
| Terminal-Bench Science 0.1 | 65% | 58% | Carbon +7 points |
Cost and speed
| Metric | Gemini 4 Carbon (High) | GPT-6.1 Sol (Max) | Difference |
|---|---|---|---|
| Cost per Intelligence Index task | $7.76 | $1.27 | Sol about 84% cheaper |
| Cost to run the Intelligence Index | $8,240 | $1,124 | Sol about 86% cheaper |
| Output tokens per task | 123K | 68K | Sol about 45% fewer |
| Reasoning tokens per task | 72K | 49K | Sol about 32% fewer |
| Output tokens to run the Index | 205M | 64M | Sol about 69% fewer |
| Output speed | 92 tokens/s | 83 tokens/s | Carbon about 11% faster |
| Time to first token | 100.31 s | 251.44 s | Carbon starts about 2.5x sooner |
| Time per task | 1342.93 s | 835.20 s | Sol finishes about 38% sooner |
GPT-6.1 Sol at every effort level
| Sol effort | Intelligence Index | Cost per task | Gap to Carbon (59) |
|---|---|---|---|
| Low | 35 | $0.16 | 24 points behind |
| Medium (default) | 46 | $0.36 | 13 points behind |
| High | 50 | $0.59 | 9 points behind |
| Xhigh | 52 | $0.76 | 7 points behind |
| Max | 55 | $1.27 | 4 points behind |
Pricing and specs
| Specification | Gemini 4 Carbon | GPT-6.1 Sol | Difference |
|---|---|---|---|
| Input, per 1M tokens | $2Introductory, then $4 | $2$4 above 272K input | Same for now |
| Cached input, per 1M tokens | $0.10Introductory, then $0.20 | $0.10 | Same for now |
| Output, per 1M tokens | $10Introductory, then $20 | $10$15 above 272K input | Same for now |
| Context window | 1M tokens | 1.05M tokens | Sol slightly larger |
| Max output | Not published | 128K tokens | - |
| Reasoning effort | Yes | Low, medium, high, xhigh, max | Sol publishes five levels |
| Input | Text, image, speech, video | Text, image | Carbon takes audio and video |
| Public API | Not yetCyber defenders only | Yesgpt-6.1-sol | Sol |
| Released | October 28, 2026 | September 29, 2026 | Carbon 29 days newer |
Smarter, but not cheaper
Carbon is ahead on most benchmarks, but Sol does the same Intelligence Index tasks for about a sixth of the cost and uses far fewer tokens to get there. Carbon's $2 and $10 is also an introductory price; after it ends, Carbon costs twice as much as Sol per token. And right now you can only use Carbon if you are one of the cyber defenders Google picked for early access, while GPT-6.1 Sol is available to every developer through OpenAI's API.