← Back to directory
Google DeepMind
Gemini 2.5 Pro
Released Mar 25, 2025 · knowledge cutoff 2025-01
- Status
- Superseded
- Location
- United States / United Kingdom
- Modality
- Multimodal
- Context window
- 1M
- Max output
- 66K
- Speed
- —
- Price ($/MTok in / out)
- $1.25 / $10
- Cost per task
- $0.20
- Open weights
- No
Benchmarks7 of 8 reported
MMLU-Pro
86.2%
GPQA Diamond
84%
SWE-bench Verified
63.8%
AIME (latest)
86.7%
Humanity's Last Exam
18.8%
LMArena Elo
1446
ARC-AGI-2
4.9%
Terminal-Bench 2.1
—
Strengths & weaknesses
Strengths
- Thinks by default, so you get deliberation without asking or configuring for it
- Takes text, images, audio and video in one prompt with no preprocessing
- Long context that holds up for whole-codebase and whole-corpus questions
Weaknesses
- Drops parts of multi-part instructions, especially formatting and negative constraints
- Drifts in agentic work — makes edits beyond what was asked
- Thinking is hard to bound; it over-deliberates on small requests
For developers
API model strings
Not researched
Licence
Proprietary — weights not released
Retirement
No retirement announced
Lineage
No recorded predecessor