B.O.B. Leaderboard
Google DeepMindreleased 28 Oct 2026
Gemini 4 Carbon
#1 of 998 of 46 benchmarksConsistent
98.8
B.O.B.
Category profile
- Real use (hard to game)2×98
- Reasoning & knowledge100
- Math—
- Coding99
- Agents & real work100
- Independent indexes—
Reality check
Lab tests
100
Real use (4 tests)
98
Lab tests and real-use tests agree. It does 2 points better on the usual lab tests than on tests that are hard to train for.
What counts as real usePrice per 1M tokens (OpenRouter)
— in · — out
Cost per task on the AA index
—
Output speed (Artificial Analysis)
—
§Receipts
Every benchmark result
“Beats” = the share of tracked models this one beat on that test. The raw score is its best reported setting.
| Benchmark | Raw | Beats | Source |
|---|---|---|---|
| Real use (hard to game) · Real use 98 | |||
| LMArena Textas Gemini 4 Carbon (High) | 1528 | 100%of 82 | LMArena (blind human votes) |
| LMArena Agent Arenaas Gemini 4 Carbon (High) | 0.160 | 98%of 49 | LMArena (real agent sessions) |
| LMArena WebDevas Gemini 4 Carbon (High) | 1768 | 97%of 87 | LMArena (blind human votes) |
| Vals Indexas google/gemini-4-carbon | 70.5% | 96%of 26 | Vals AI (private test sets) |
| Reasoning & knowledge · Reasoning 100 | |||
| CritPt (physics research)as gemini-4-carbon_high | 36.0% | 100%of 98 | Epoch AI Benchmarking Hub |
| Blueprint-Bench 2as gemini-4-carbon_unknown | 58.6% | 100%of 33 | Epoch AI Benchmarking Hub |
| Coding · Coding 99 | |||
| SciCodeas gemini-4-carbon_high | 64.2% | 99%of 90 | Epoch AI Benchmarking Hub |
| Agents & real work · Agents 100 | |||
| Vending-Bench 2as gemini-4-carbon_unknown | 16484.2% | 100%of 49 | Epoch AI Benchmarking Hub |