ClosedGoogle DeepMind/10 trustVerified

Gemini 2.5 Pro

Name: Gemini 2.5 Pro
Rating: 8

Google's flagship 2025 reasoning model — 1M-token context, top of the leaderboards.

Visit official page →

About

Gemini 2.5 Pro is Google DeepMind's reasoning-focused flagship from the 2.5 generation,

released as an experimental preview on March 25, 2025 (general availability mid-June 2025). It is a

"thinking" model by default — it reasons step-by-step internally before answering — and handles text,

image, audio, and video inputs in a single 1,048,576-token (1M) context window. A June 2025 upgraded

preview pushed it to the top of the LMArena leaderboard (1470 Elo) and WebDevArena (1443 Elo), and led

the Aider Polyglot coding benchmark (all Google-reported, with LMArena independently observed via

human-preference voting).

Provider-reported benchmark highlights (Google launch materials):

Humanity's Last Exam (no tools): 18.8% — state of the art at launch
GPQA Diamond: 84%
AIME 2025: 86.7%
SWE-bench Verified: 63.8% (2nd to Claude 3.7 Sonnet at the time)

Available via the Gemini app,

Google AI Studio, the Gemini Developer API, and Vertex AI, with a free tier for rate-limited use.

Specs

Context window

1,048,576 tokens (1M)

Pricing

$1.25 / $10.00 per 1M tokens in/out for prompts ≤200k tokens (verified against Google's official pricing page, ai.google.dev); $2.50 / $15.00 per 1M tokens for prompts >200k tokens (rate applies to the entire prompt). Batch API ~50% off standard rates.

Free quota

Free tier in Google AI Studio / Gemini API (rate-limited, no credit card); full access free for Gemini Advanced subscribers.

API access

Gemini Developer API (REST/SDK via Google AI Studio), Vertex AI (GCP), and the Gemini app.

Weights

Closed — API only

Released

March 25, 2025

Benchmarks

Provider-reported (Google): Humanity's Last Exam 18.8% (no tools, SOTA at launch); GPQA Diamond 84%; AIME 2025 86.7%; SWE-bench Verified 63.8%.

Independent: #1 on LMArena at launch (1470 Elo after the June 2025 upgrade, human-preference votes); #1 on WebDevArena (1443 Elo, Google-reported from public arena data); covered by Artificial Analysis benchmarks.

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

complex reasoningcoding agentslong-context document analysismultimodal analysismath and science

Trust score — 8/10

/10 trustVerified

Scored 8/10 (verified tier, metrics-v1). Evidence: provider-tier estimate — thinner evidence (no public adoption metric); Google flagship; no public adoption metric; released 567 days ago. Manual review: pricing verified on official pricing page; #1 LMArena (independent); provider-reported benchmarks labeled. License: closed weights.

How the trust score is built

Full methodology →

Adoption · stars/downloads, log-scaled vs peers: 10.0/10
Maintenance · recency, releases, issue load: 5.0/10
Openness · license + docs: 5.1/10
Manual review · sanity read + red-flag check: 9.0/10

· provider-tier estimate — thinner evidence (no public adoption metric)
· Google flagship; no public adoption metric
· released 567 days ago
· manual review: pricing verified on official pricing page; #1 LMArena (independent); provider-reported benchmarks labeled