GoogleReasoning model

Gemini 3.5 Flash

Reasoning effort

Gemini 3.5 Flash is a Google proprietary reasoning model, released 19 May 2026. Its best configuration (medium reasoning effort) ranks #68 of 751 on the BenchLeader Index at 63.1 ±2.6, in the upper half. The ± is the point: 131 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in instruction following (73) and lowest in coding (57). At $3.38 per million tokens blended it is among the most expensive ranked models. Output speed of 211 tokens per second puts it in the fastest quarter, with a first answer in 14.2 s. It has been measured at 5 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 2 Oct 2026.

Blended price: $3.38/M; $1.50 in · $9.00 out
Output speed: 211 tok/s; measured by Artificial Analysis
First answer: 14 s; first token 1.63 s
Context: 1.0M
Full answer: 17 s; median, reasoning included
Cost per run: –
Released: 19 May 2026

How it scores by categoryDashed line = average model (50). One step of 15 = one standard deviation.

Overall index63
Reasoning62
Coding57
Agents & tools68
Maths67
Knowledge71
Instruction following73
Human preference67
Multimodal67
Long context62
Composite68

Versions

Google has shipped 13 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.

Model	Released	Index	Rank
Gemini 3.8 Flashmedium	2 Sept 2026	64.7	#48
Gemini 3.7 Flashmedium	13 Aug 2026	63.5	#63
Gemini 3.6 Flashhigh	21 Jul 2026	60.5	#102
Gemini 3.5 Flashmediumthis page	19 May 2026	63.1	#68
Gemini 3 Flashthinking	17 Dec 2025	59.3	#122
Gemini 2.5 Flash 09 2025	25 Sept 2025	51.9	#314
Gemini 2.5 Flash	17 Apr 2025	52.8	#283
Gemini 2.0 Flashthinking	5 Feb 2025	45.0	#496
Gemini 1.5 Flash 002	24 Sept 2024	40.7	#623
Gemini 1.5 Flash 001	23 May 2024	36.4	#717
Gemini 1.5 Flash 0514	14 May 2024	–	–
Gemini 1.5 Flash	–	–	–
Gemini Flash 2.0	–	–	–

Reasoning-effort configurations

The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.

Effort	Index	Rank	Speed	First answer	Chat reply cost	Categories
minimal	56.5	#191	198 tok/s	0.98 s	$0.0033	Agents & tools 74 · Composite 57 · Instruction following 49 · Knowledge 62 · Long context 56 · Maths 59 · Multimodal 63 · Reasoning 47
low	–	–	87 tok/s	1.63 s	$0.0033	Maths 63 · Reasoning 64
mediumbest	63.1	#68	211 tok/s	14 s	$0.0033	Agents & tools 68 · Coding 57 · Composite 68 · Human preference 67 · Instruction following 73 · Knowledge 71 · Long context 62 · Maths 67 · Multimodal 67 · Reasoning 62
high	62.6	#78	216 tok/s	16 s	$0.0033	Agents & tools 63 · Coding 60 · Composite 58 · Human preference 67 · Instruction following 74 · Knowledge 65 · Long context 62 · Maths 59 · Multimodal 67 · Reasoning 62
not stated	49.6	#371	87 tok/s	1.63 s	$0.0033	Agents & tools 35 · Human preference 33 · Maths 58 · Reasoning 72

How its index has moved

8 Sept 2026 to 3 Oct 2026

medium
high
minimal
not stated

The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.

Benchmark results

One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.

Reasoning

Benchmark	minimal	low	medium	high	not stated	Source
GPQA Diamond	86.4%#72	88.9%#51	–	92.8%#22	–	Epoch AI Benchmarking Hub
SimpleBench	–	–	–	–	76.7%#6	SimpleBench
LMArena Hard Prompts	–	–	1495#42	1495#42	–	LMArena
LiveBench Reasoningnot in index	–	–	–	82.0%#45	–	LiveBench
GPQA Diamond (AA)not in index	82.8%#162	–	92.1%#39	92.2%#38	–	Artificial Analysis
Humanity's Last Exam (AA)not in index	24.1%#170	–	41.3%#69	42.7%#57	–	Artificial Analysis
GPQA Diamond (Vals)not in index	–	–	–	92.7%#14	–	Vals AI
EnigmaEval	–	–	–	25.4%#4	–	Scale AI SEAL
ARC-AGI-1	48.8%#150	–	–	92.5%#48	–	ARC Prize
ARC-AGI-2	8.9%#137	–	–	72.1%#49	–	ARC Prize
CritPt	1.4%#190	–	10.9%#101	13.1%#92	–	Artificial Analysis
Chess Puzzlesnot in index	43.0%#19	45.0%#17	–	50.0%#11	–	Epoch AI Benchmarking Hub
Mystery Game Puzzlesnot in index	–	28.0%#34	–	32.0%#26	–	Epoch AI Benchmarking Hub
LMCAnot in index	–	–	–	47.1%#63	–	Epoch AI Benchmarking Hub
DTBenchnot in index	–	–	–	94.7%#39	–	Epoch AI Benchmarking Hub
ForecastBenchnot in index	–	–	–	–	59.0%#45	ForecastBench

Coding

Benchmark	minimal	low	medium	high	not stated	Source
SWE-bench Verified (Epoch)	–	–	–	79.3%#3	–	Epoch AI Benchmarking Hub
SciCode	–	–	–	53.1%#65	–	SciCode
WeirdML	–	–	–	62.6%#43	–	WeirdML
LMArena Coding	–	–	1506#65	1507#63	–	LMArena
LMArena WebDev	–	–	1490#54	1499#52	–	LMArena
LiveBench Codingnot in index	–	–	–	78.2%#31	–	LiveBench
SciCode (AA)not in index	–	–	–	53.9%#65	–	Artificial Analysis
LiveCodeBench	–	–	–	87.6%#12	–	Vals AI
SWE-bench (Vals)not in index	–	–	–	78.8%#28	–	Vals AI
Code Migrationnot in index	–	–	–	26.8%#38	–	Vals AI
ProgramBenchnot in index	–	–	–	0.0%#23	–	Vals AI
Vibe Code Bench v1.1not in index	–	–	–	48.7%#49	–	Vals AI
Surface Evolver Benchnot in index	–	–	58.1%#14	–	–	Epoch AI Benchmarking Hub
DeepSWEnot in index	–	–	37.4%#58	36.1%#60	–	Epoch AI Benchmarking Hub
ALE-Benchnot in index	–	–	–	911.0#55	–	Epoch AI Benchmarking Hub
Terminal-Bench 4.0 (AA)not in index	–	–	–	6.6%#84	–	Artificial Analysis
Terminal-Bench 2.1 (AA)not in index	–	–	–	78.7%#49	–	Artificial Analysis

Agents & tools

Benchmark	minimal	low	medium	high	not stated	Source
APEX-Agents	–	–	–	–	27.5%#36	Mercor
LiveBench Agentic Codingnot in index	–	–	–	49.0%#42	–	LiveBench
Terminal-Bench Hard	46.2%#30	–	39.4%#58	40.9%#53	–	Artificial Analysis
τ²-Bench Telecom (AA)not in index	58.8%#182	–	95.6%#20	95.3%#24	–	Artificial Analysis
Terminal-Bench 2.1 (Vals)	–	–	–	74.2%#20	–	Vals AI
MCP Atlas	–	–	–	83.6%#6	–	Scale AI SEAL
HiL-Bench	–	–	–	–	27.7%#12	Scale AI SEAL
GDPval (AA)not in index	–	–	–	34.2%#109	–	Artificial Analysis
τ²-Bench Banking (AA)	–	–	–	32.2%#57	–	Artificial Analysis
ITBench SRE (AA)not in index	–	–	–	40.4%#17	–	Artificial Analysis
Analyst Agent (AA)not in index	–	–	–	45.0%#12	–	Artificial Analysis
APEX-Agents (AA)not in index	–	–	–	47.0%#1	–	Artificial Analysis
Finance Agent v2not in index	–	–	–	57.9%#11	–	Vals AI
Harvey's Legal Agent Benchmarknot in index	–	–	–	2.5%#38	–	Vals AI
Legal Research Benchnot in index	–	–	–	30.8%#37	–	Vals AI
SkillsBenchnot in index	–	–	–	52.7%#18	–	Vals AI
Tax Agent Benchnot in index	–	–	–	53.5%#37	–	Vals AI
Terminal-Bench 2.0 (Vals)not in index	–	–	–	67.4%#4	–	Vals AI
Terminal-Bench 4.0 (Vals)not in index	–	–	–	6.1%#32	–	Vals AI
Terminal-Bench Sciencenot in index	–	–	–	5.7%#16	–	Vals AI
Web Search Indexnot in index	–	–	–	37.0%#7	–	Vals AI
GBAEvalnot in index	–	–	–	–	6.7%#14	Epoch AI Benchmarking Hub
Vending-Bench 2not in index	–	–	–	–	5396.4#23	Andon Labs
AutomationBenchnot in index	–	–	–	42.1%#39	–	Artificial Analysis
GDP.pdfnot in index	–	–	–	19.8%#29	–	Artificial Analysis
Harvey LABnot in index	–	–	–	82.1%#16	–	Harvey
EnterpriseOps-Gymnot in index	–	–	–	50.1%#3	–	Artificial Analysis
AA-Briefcasenot in index	–	–	–	873#42	–	Artificial Analysis
FACTS Searchnot in index	–	–	–	–	64.3%#16	Google

Maths

Benchmark	minimal	low	medium	high	not stated	Source
FrontierMath Tiers 1–3	–	–	–	62.8%#33	–	Epoch AI Benchmarking Hub
FrontierMath Tier 4	–	–	–	26.8%#38	–	Epoch AI Benchmarking Hub
OTIS Mock AIME	80.0%#106	88.9%#68	–	95.6%#36	–	Epoch AI Benchmarking Hub
ProofBench	–	–	–	31.0%#35	–	Vals AI
LiveBench Mathematicsnot in index	–	–	–	88.2%#40	–	LiveBench
AIME 2026	–	–	–	–	95.0%#18	MathArena
HMMT February 2026	–	–	–	–	95.5%#6	MathArena
MathArena Apex	–	–	–	–	32.3%#9	MathArena
LMArena Maths	–	–	1482#32	1497#19	–	LMArena

Knowledge

Benchmark	minimal	low	medium	high	not stated	Source
SimpleQA Verified	–	–	–	66.2%#11	–	Epoch AI Benchmarking Hub
LiveBench Data Analysisnot in index	–	–	–	64.9%#56	–	LiveBench
AA-Omniscience	0.7#118	–	20.8#61	21.2#58	–	Artificial Analysis
MMLU-Pro	–	–	–	89.5%#10	–	Vals AI
LegalBench	–	–	–	83.6%#49	–	Vals AI
CorpFin	–	–	–	64.7%#35	–	Vals AI
TaxEval	–	–	–	74.4%#36	–	Vals AI
Excel Modeling Benchmarknot in index	–	–	–	63.5%#25	–	Vals AI
MedCodenot in index	–	–	–	55.8%#6	–	Vals AI
MedScribenot in index	–	–	–	76.6%#70	–	Vals AI
Public Benefits Benchnot in index	–	–	–	59.5%#28	–	Vals AI
DrugDiscoveryBenchnot in index	–	–	–	50.0%#7	–	Scale AI SEAL
EBR-benchnot in index	–	–	–	4.8%#22	–	Epoch AI Benchmarking Hub
GDP.pdfnot in index	–	–	14.0%#35	–	14.0%#35	Epoch AI Benchmarking Hub
AA-Omniscience: accuracynot in index	43.0%#99	–	51.0%#58	51.4%#56	–	Artificial Analysis
AA-Omniscience: non-hallucinationnot in index	25.7%#223	–	38.2%#157	37.8%#158	–	Artificial Analysis

Instruction following

Benchmark	minimal	low	medium	high	not stated	Source
LiveBench Languagenot in index	–	–	–	84.6%#16	–	LiveBench
LiveBench Instruction Followingnot in index	–	–	–	75.6%#7	–	LiveBench
IFBench	47.3%#187	–	74.6%#35	76.3%#21	–	Artificial Analysis
LMArena Instruction Followingnot in index	–	–	1466#41	1465#43	–	LMArena

Human preference

Benchmark	minimal	low	medium	high	not stated	Source
LMArena Text	–	–	1476#31	1477#28	–	LMArena
EQ-Bench 4	–	–	–	–	1087#23	EQ-Bench
LMArena Creative Writingnot in index	–	–	1469#18	1464#22	–	LMArena
LMArena Multi-turnnot in index	–	–	1478#44	1481#39	–	LMArena
LMArena Longer Queriesnot in index	–	–	1481#46	1482#39	–	LMArena

Multimodal

Benchmark	minimal	low	medium	high	not stated	Source
LMArena Vision	–	–	1308#13	1310#12	–	LMArena
MMMU-Pro	80.1%#50	–	83.9%#24	84.3%#20	–	Artificial Analysis
MMMU-Pro (Vals)not in index	–	–	–	88.3%#8	–	Vals AI
MortgageTaxnot in index	–	–	–	68.1%#19	–	Vals AI
SAGEnot in index	–	–	–	49.9%#18	–	Vals AI
Blueprint-Bench 2not in index	–	–	–	–	33.6%#7	Epoch AI Benchmarking Hub

Long context

Benchmark	minimal	low	medium	high	not stated	Source
AA-LCR	61.3%#254	–	74.3%#148	73.3%#164	–	Artificial Analysis
LMArena Documentnot in index	–	–	1461#25	1453#28	–	LMArena
MLCRnot in index	–	–	–	18.3%#20	–	Artificial Analysis

Composite

Benchmark	minimal	low	medium	high	not stated	Source
Epoch Capabilities Indexnot in index	–	–	–	–	154.5#32	Epoch AI Benchmarking Hub
LiveBench	–	–	–	74.6%#38	–	LiveBench
AA Intelligence Index	23.9#166	–	33.6#92	32.6#97	–	Artificial Analysis
Vals Indexnot in index	–	–	–	44.8#31	–	Vals AI
Vals Multimodal Indexnot in index	–	–	–	62.9%#16	–	Vals AI

Where to run it

Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.

Provider	Speed	First token	Input $/M	Output $/M	Context	Quantisation
Google AI Studio	175 tok/s	1.80 s	$1.50	$9.00	1.0M	–
Google AI Studio	127 tok/s	1.15 s	$0.750	$4.50	1.0M	–
Google Vertex	122 tok/s	1.58 s	$1.50	$9.00	1.0M	–
Google Vertex	103 tok/s	13 s	$0.750	$4.50	1.0M	–
Google AI Studio	78 tok/s	1.01 s	$2.70	$16.20	1.0M	–
Google Vertex	69 tok/s	1.08 s	$2.70	$16.20	1.0M	–

What a task costs

Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache at $0.150 per 1M. Reasoning tokens are not modelled.

Workload	Tokens in / out	Cost	With caching	Time
Chat reply	400 / 300	$0.0033	$0.0029	15.6 s
Summarise a 30-page report	12,000 / 600	$0.023	$0.011	17.0 s
Code edit	6,000 / 1,500	$0.022	$0.016	21.3 s
Agentic coding session	60,000 / 4,000	$0.126	$0.065	33.1 s
Structured extraction	2,000 / 200	$0.0048	$0.0028	15.1 s