A · closed · Gemini

Gemini 2.5 Flash-Lite

Google

B · closed · Gemini

Gemini 3.5 Flash-Lite

Google

Replace A

Replace B

0:7

overall capability wins (1 ties across 8 scored). Gemini 3.5 Flash-Lite leads overall.

Reasoning0:5
Coding0:1
Pricing3:0
Speed2:0
Specs0:1

Verdict

Gemini 3.5 Flash-Lite leads coding (widest gap: +22.0 on SciCode); Gemini 3.5 Flash-Lite offers larger context (1.0M vs 1M); Gemini 2.5 Flash-Lite is ~4.9x cheaper on a blended token basis than Gemini 3.5 Flash-Lite.

Gemini 3.5 Flash-Lite leads coding (widest gap: +22.0 on SciCode)
Gemini 3.5 Flash-Lite offers larger context (1.0M vs 1M)
Gemini 2.5 Flash-Lite is ~4.9x cheaper on a blended token basis than Gemini 3.5 Flash-Lite

Cost for 1M in + 250K out

Illustrative chat workload at primary-provider list prices.

Gemini 2.5 Flash-Lite: $0.20
Gemini 3.5 Flash-Lite: $0.93

Gemini 2.5 Flash-Lite is about 4.6x cheaper on this workload.

Price per 1M tokens

Full comparison

Organization

Google

Family

Gemini

License

Proprietary

Open weights

Release

Jun 17, 2025

Jul 21, 2026

Knowledge cutoff

API / provider

Google

Modalities

text, image, audio, video → text

Specs

Context window

1.0M

B +49K

Max output

66K

tie

Parameters

—

Pricing

Input $/1M

$0.10

$0.30

A +$0.20

Output $/1M

$0.40

$2.50

A +$2.10

Blended $/1M (3∶1)

$0.18

$0.85

A +$0.67

Speed

tok/s

250

220

A +30

TTFT (s)

0.12

0.18

A +0.06

Reasoning

MMLU-Pro

—

GPQA Diamond

62.5

83.8

B +21.3 pts

Humanity's Last Exam

6.8

18.8

B +12.0 pts

AIME 2025

—

MATH-500

—

Humanity's Last Exam (with tools)

—

AA-Omniscience Accuracy

17.9

29.5

B +11.6 pts

AA-LCR v1.1

55.7

76.0

B +20.3 pts

CritPt

—

MMMU-Pro

58.2

79.0

B +20.8 pts

IFBench

49.9

—

Chartography

—

Chartography (With Tools)

—

Coding

SWE-bench Verified

—

SWE-bench Pro

—

SWE-bench Multilingual

—

LiveCodeBench

—

Terminal-Bench 2.1

—

53.6

—

Aider Polyglot

—

Terminal-Bench 3

—

BigCodeBench

—

SciCode

19.3

41.3

B +22.0 pts

CursorBench

—

SWE-Rebench

—

NL2Repo-Bench

—

DeepSWE

—

WebDev Arena

—

1,447

—

Terminal-Bench 4.0

—

CursorBench 4.0

—

FrontierCode v1.1 (Main)

—

Arena

LMArena Elo

—

1,456

—

Metric	Gemini 2.5 Flash-Lite	Gemini 3.5 Flash-Lite	Delta
Identity
Organization	Google	Google	—
Family	Gemini	Gemini	—
License	Proprietary	Proprietary	—
Open weights	No	No	—
Release	Jun 17, 2025	Jul 21, 2026	—
Knowledge cutoff	-	-	—
API / provider	Google	Google	—
Modalities	text, image, audio, video → text	text, image, audio, video → text	—
Specs
Context window	1M	1.0M	B +49K
Max output	66K	66K	tie
Parameters	—	—	—
Pricing
Input $/1M	$0.10	$0.30	A +$0.20
Output $/1M	$0.40	$2.50	A +$2.10
Blended $/1M (3∶1)	$0.18	$0.85	A +$0.67
Speed
tok/s	250	220	A +30
TTFT (s)	0.12	0.18	A +0.06
Reasoning
MMLU-Pro	—	—	—
GPQA Diamond	62.5	83.8	B +21.3 pts
Humanity's Last Exam	6.8	18.8	B +12.0 pts
AIME 2025	—	—	—
MATH-500	—	—	—
Humanity's Last Exam (with tools)	—	—	—
AA-Omniscience Accuracy	17.9	29.5	B +11.6 pts
AA-LCR v1.1	55.7	76.0	B +20.3 pts
CritPt	—	—	—
MMMU-Pro	58.2	79.0	B +20.8 pts
IFBench	49.9	—	—
Chartography	—	—	—
Chartography (With Tools)	—	—	—
Coding
SWE-bench Verified	—	—	—
SWE-bench Pro	—	—	—
SWE-bench Multilingual	—	—	—
LiveCodeBench	—	—	—
Terminal-Bench 2.1	—	53.6	—
Aider Polyglot	—	—	—
Terminal-Bench 3	—	—	—
BigCodeBench	—	—	—
SciCode	19.3	41.3	B +22.0 pts
CursorBench	—	—	—
SWE-Rebench	—	—	—
NL2Repo-Bench	—	—	—
DeepSWE	—	—	—
WebDev Arena	—	1,447	—
Terminal-Bench 4.0	—	—	—
CursorBench 4.0	—	—	—
FrontierCode v1.1 (Main)	—	—	—
Arena
LMArena Elo	—	1,456	—