ClosedGoogle/10 trustVerified

Gemini 2.5 Flash

Name: Gemini 2.5 Flash
Rating: 8

Google's hybrid-reasoning workhorse: 1M-token multimodal context at a mid-tier price.

Visit official page →

About

Gemini 2.5 Flash is Google's mid-tier model and its first fully hybrid reasoning model: developers can toggle internal "thinking" on or off per request or cap it at a 0-24,576 token thinking budget, tuning cost against quality on the fly. It is a sparse mixture-of-experts transformer with native multimodal input (text, images, video, audio) and text output, a 1M-token input context with 64K output, and a January 2025 knowledge cutoff.

Benchmarks below are provider-reported from Google's Gemini 2.5 Flash model card (GA build, thinking mode, pass@1): GPQA Diamond 82.8%, SWE-bench Verified 60.4%, AIME 2025 72.0%, MMMU 79.7%, Global MMLU Lite 88.4% — strong on science/math, noticeably behind Claude-tier coders on software engineering. Independent long-context checks flag degradation: MRCR v2 retrieval falls from 74% at 128K to ~32% at the full 1M window (third-party-reported).

Specs

Context window

1,048,576 tokens input / 65,536 tokens output

Pricing

$0.30/M input (text/image/video/audio), $2.50/M output (Standard tier) — verified on Google's pricing page 2026-10-13; Batch tier $0.15/$1.25; context caching $0.03/M + $1.00/M tokens/hour storage

Free quota

Free tier in Google AI Studio (rate-limited; exact RPD limits not published on the pricing page). Free-tier content may be used to improve Google's products.

API access

Gemini API, Google AI Studio, Vertex AI

Weights

Closed — API only

Released

June 17, 2025

Benchmarks

Provider-reported (Google model card, GA build, thinking mode):

GPQA Diamond: 82.8%
SWE-bench Verified: 60.4%
AIME 2025: 72.0%
MMMU: 79.7%
Global MMLU Lite: 88.4%
MRCR v2 (third-party-reported long-context retrieval): 74% at 128K, 32% at 1M

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

Use cases

high-volume agentic workloadsmultimodal classification and summarizationvideo and audio transcription/pipelineslong-context chatbots and RAGcost-sensitive reasoning tasks with a thinking budget

Trust score — 8/10

/10 trustVerified

Scored 8/10 (verified tier, metrics-v1). Evidence: provider-tier estimate — thinner evidence (no public adoption metric); Google model; no public adoption metric; released 483 days ago. Manual review: pricing verified on official pricing page; Oct-2026-shutdown rumor is unverified third-party claim (Google deprecation page lists no date); benchmarks provider-reported. 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: 8.0/10

· provider-tier estimate — thinner evidence (no public adoption metric)
· Google model; no public adoption metric
· released 483 days ago
· manual review: pricing verified on official pricing page; Oct-2026-shutdown rumor is unverified third-party claim (Google deprecation page lists no date); benchmarks provider-reported