Gemini 4 Carbon
Google's most capable reasoning model for complex software engineering, deep math and science, high factual accuracy, and long-horizon autonomous agents
Shogo Agent Index
Supported#2 of 9273.6range 67.2–80 · 27% of weight measured
Agentic coding
82
26% measured
Long-horizon autonomy
80
50% measured
Knowledge-work agents
63
32% measured
Computer use
68
estimated
Security & context learning
63
estimated
Price
- Input
- $2.00/M
- Output
- $10.00/M
- Cache read
- $0.10/M
- 3:1 blend
- $4.00/M
- Agent-effective
- $0.84/M
Cheapest host for agents: Google AI Studio at $0.42/M.
Agent-effective = 20 input tokens per output token, 85% cache hits. Why.
Agentic benchmarks
| Benchmark | Score | Normalised | Index weight | Harness / source |
|---|---|---|---|---|
| SWE-bench Verified | est. | 8% | predicted 90 | |
| Terminal-Bench 2.0 | est. | 8% | predicted 97 | |
| DeepSWE | est. | 6% | predicted 84 | |
| FrontierSWE | 61.2% | 5% | proximus · $61.93/task | |
| CursorBench | 60.4% | 4% | source · $9.22/task | |
| GSO-Bench | est. | 4% | predicted 80 | |
| METR Time Horizons | est. | 10% | predicted 87 | |
| Vending-Bench 2 | $16,484 | 6% | source | |
| PostTrainBench | 41.0% | 4% | OpenCode | |
| APEX-Agents | 84.2% | 8% | source | |
| GDPval | est. | 7% | predicted 73 | |
| Remote Labor Index | est. | 5% | predicted 24 | |
| DeepResearch Bench | est. | 5% | predicted 53 | |
| OSWorld 2.0 | est. | 6% | predicted 50 | |
| OSWorld | est. | 4% | predicted 96 | |
| Cybench | est. | 5% | predicted 95 | |
| CL-bench | est. | 5% | predicted 31 |
Other benchmarks
| Benchmark | Category | Score | Source |
|---|---|---|---|
| ARC-AGI | Reasoning | 98.5% | source |
| ARC-AGI-2 | Reasoning | 94.2% | source |
| Chess Puzzles | Reasoning | 78.0% | source |
| CritPt | Reasoning | 36.0% | source |
| FrontierMath Tiers 1-3 | Math | 90.5% | source |
| FrontierMath Tier 4 | Math | 95.1% | source |
| GPQA Diamond | Reasoning | 96.5% | source |
| OTIS Mock AIME | Math | 100.0% | source |
| SciCode | Coding | 64.2% | source |
| SimpleBench | Reasoning | 92.6% | source |
| SimpleQA Verified | Knowledge | 77.7% | source |
API hosts
Live prices and uptime for every host serving Gemini 4 Carbon via OpenRouter.
| Host | Input | Output | Cache read | Agent $/M | Uptime 24h | Quant |
|---|---|---|---|---|---|---|
| Google AI Studio | $1.00 | $5.00 | $0.050 | $0.42 | 100.0% | — |
| $2.00 | $10.00 | $0.10 | $0.84 | 52.6% | — | |
| Google AI Studio | $2.00 | $10.00 | $0.10 | $0.84 | 99.9% | — |
| $3.60 | $18.00 | $0.18 | $1.52 | 58.7% | — | |
| Google AI Studio | $3.60 | $18.00 | $0.18 | $1.52 | 92.3% | — |