GoogleReasoning modelCompare with another model

Gemini 3.8 Flash

Reasoning effort

Gemini 3.8 Flash is a Google proprietary reasoning model, released 2 Sept 2026. Its best configuration (high reasoning effort) ranks #48 of 775 on the BenchLeader Index at 64.5 ±3.9, in the upper half. The ± is the point: 125 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 composite (75) and lowest in agents & tools (51). At $1.50 per million tokens blended it is pricier than most ranked models. Output speed of 144 tokens per second puts it faster than most, with a first answer in 26.2 s. It has been measured at 4 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 19 Oct 2026.

Blended price: $1.50/M; $0.750 in · $3.75 out
Output speed: 144 tok/s; measured by Artificial Analysis
First answer: 26 s; first token 2.50 s
Context: 1.0M
Full answer: 30 s; median, reasoning included
Cost per run: $1.24; one full Intelligence Index run
Released: 2 Sept 2026

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

Overall index65
Reasoning68
Coding64
Agents & tools51
Maths63
Knowledge68
Human preference70
Multimodal68
Long context66
Composite75

Versions

Google has shipped 12 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 Flashhighthis page	2 Sept 2026	64.5	#48
Gemini 3.7 Flashhigh	13 Aug 2026	63.5	#61
Gemini 3.6 Flashhigh	21 Jul 2026	62.1	#89
Gemini 3.5 Flashhigh	19 May 2026	63.2	#68
Gemini 3 Flashthinking	17 Dec 2025	59.2	#134
Gemini 2.5 Flash 09 2025thinking	25 Sept 2025	51.1	#348
Gemini 2.5 Flash	17 Apr 2025	52.7	#302
Gemini 2.0 Flash	5 Feb 2025	43.3	#560
Gemini 1.5 Flash 002	24 Sept 2024	40.2	#650
Gemini 1.5 Flash 001	23 May 2024	37.2	#726
Gemini 1.5 Flash	14 May 2024	–	–
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
low	61.1	#102	77 tok/s	2.50 s	$0.0014	Agents & tools 62 · Coding 61 · Composite 67 · Knowledge 71 · Long context 65 · Multimodal 68 · Reasoning 57
medium	64.5	#49	77 tok/s	2.50 s	$0.0014	Agents & tools 75 · Coding 58 · Composite 74 · Knowledge 74 · Long context 67 · Multimodal 67 · Reasoning 63
highbest	64.5	#48	144 tok/s	26 s	$0.0014	Agents & tools 51 · Coding 64 · Composite 75 · Human preference 70 · Knowledge 68 · Long context 66 · Maths 63 · Multimodal 68 · Reasoning 68
not stated	55.6	#223	77 tok/s	2.50 s	$0.0014	Agents & tools 53 · Composite 49 · Knowledge 58 · Maths 53 · Reasoning 75

How its index has moved

8 Sept 2026 to 21 Oct 2026

medium
high
low
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

See all LLM benchmarks. 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	low	medium	high	not stated	Source
GPQA Diamond	–	–	95.4%#4	–	Epoch AI Benchmarking Hub
Humanity's Last Exameffort not stated	–	–	–	44.5%#5	Scale AI / CAIS
Arena Hard Prompts	–	–	1518#10	–	Arena (LMArena)
LiveBench 2.0 Reasoningnot in indexeffort not stated	–	–	–	89.3%#17	LiveBench
LiveBench 1.0 Reasoningnot in index	–	–	89.3%#18	–	LiveBench
GPQA Diamond (AA)not in index	92.0%#41	93.5%#14	95.3%#3	–	Artificial Analysis
Humanity's Last Exam (AA)not in index	37.1%#106	42.1%#73	47.8%#37	–	Artificial Analysis
GPQA Diamond (Vals)not in index	–	–	94.4%#4	–	Vals AI
ARC-AGI-1	90.5%#68	97.5%#12	98.5%#1	–	ARC Prize
ARC-AGI-2	77.5%#42	82.9%#38	89.2%#18	–	ARC Prize
ARC-AGI-3	15.1%#19	24.2%#17	35.0%#15	–	ARC Prize
CritPt	4.0%#165	12.3%#106	18.3%#69	–	Artificial Analysis
MysteryMechanismnot in index	–	–	36.5%#8	–	Vals AI
Chess Puzzlesnot in index	–	–	58.0%#14	–	Epoch AI Benchmarking Hub
Mystery Game Puzzlesnot in index	–	–	47.0%#14	–	Epoch AI Benchmarking Hub
LMCAnot in index	51.2%#61	52.9%#51	–	51.5%#58	Epoch AI Benchmarking Hub
DTBenchnot in index	94.9%#46	94.9%#46	–	95.7%#37	Epoch AI Benchmarking Hub

Coding

Benchmark	low	medium	high	not stated	Source
SciCode	55.0%#52	55.1%#50	56.6%#35	–	SciCode
WeirdML	–	–	84.8%#15	–	WeirdML
FrontierCode	–	41.2%#25	–	–	Cognition
Arena Coding	–	–	1528#26	–	Arena (LMArena)
Arena WebDev	–	–	1585#32	–	Arena (LMArena)
LiveBench 2.0 Codingnot in indexeffort not stated	–	–	–	82.2%#28	LiveBench
LiveBench 1.0 Codingnot in index	–	–	72.5%#57	–	LiveBench
SciCode (AA)not in index	55.0%#54	55.1%#52	56.6%#35	–	Artificial Analysis
LiveCodeBench	–	–	89.5%#3	–	Vals AI
IOI	–	–	56.9%#18	–	Vals AI
SWE-bench (Vals)not in index	–	–	80.0%#24	–	Vals AI
Code Migrationnot in index	–	–	36.5%#35	–	Vals AI
ProgramBenchnot in index	–	–	1.0%#15	–	Vals AI
Vibe Code Bench 1-100not in index	–	–	18.8%#13	–	Vals AI
Vibe Code Bench v1.1not in index	–	–	78.7%#28	–	Vals AI
SWE Atlas: Codebase QnAnot in indexeffort not stated	–	–	–	47.0%#6	Scale AI SEAL
SWE Atlas: Refactoringnot in indexeffort not stated	–	–	–	44.8%#7	Scale AI SEAL
SWE Atlas: Test Writingnot in indexeffort not stated	–	–	–	53.7%#4	Scale AI SEAL
Surface Evolver Benchnot in index	–	–	76.9%#9	–	Epoch AI Benchmarking Hub
DeepSWE v1.1not in index	–	71.0%#10	73.8%#2	–	Epoch AI Benchmarking Hub
CursorBenchnot in index	–	37.3%#45	39.6%#40	–	Cursor
ALE-Benchnot in index	–	–	1269.7#36	–	Epoch AI Benchmarking Hub
FrontierSWEnot in index	–	–	19.6%#16	–	Epoch AI Benchmarking Hub
Terminal-Bench 4.0 (AA)not in index	10.1%#87	19.7%#63	19.7%#63	–	Artificial Analysis
Terminal-Bench 2.1 (AA)not in index	83.2%#36	83.9%#34	87.6%#18	–	Artificial Analysis
Startrise Frontendnot in indexeffort not stated	–	–	–	64.6%#26	Startrise
FrontierCode 1.1 (AA)not in index	–	–	36.4%#31	–	Artificial Analysis

Agents & tools

Benchmark	low	medium	high	not stated	Source
Terminal-Bench	–	–	19.1%#72	–	Terminal-Bench
Remote Labor Indexeffort not stated	–	–	–	5.8%#7	Scale AI / CAIS
APEX-Agentseffort not stated	–	–	–	64.3%#9	Mercor
Arena Agent	–	–	3.7#13	–	Arena (LMArena)
LiveBench Long Agenticnot in indexeffort not stated	–	–	–	30.1%#31	LiveBench
LiveBench Browsernot in indexeffort not stated	–	–	–	64.0%#25	LiveBench
LiveBench Agentic Codingnot in index	–	–	54.2%#33	–	LiveBench
Terminal-Bench 2.1 (Vals)	–	–	81.3%#8	–	Vals AI
HiL-Bencheffort not stated	–	–	–	41.5%#16	Scale AI SEAL
GDPval-AA v2.1not in index	40.4%#101	46.1%#70	46.8%#65	–	Artificial Analysis
τ³-Banking (AA)	33.2%#56	45.8%#12	45.0%#15	–	Artificial Analysis
ITBench SRE (AA)not in index	–	–	52.5%#3	–	Artificial Analysis
Analyst Agent (AA)not in index	–	–	46.3%#15	–	Artificial Analysis
CUA-benchnot in index	–	–	4.2%#9	–	Vals AI
CyberBenchnot in index	–	–	43.8%#40	–	Vals AI
Finance Agent v2not in index	–	–	61.4%#2	–	Vals AI
Harvey's Legal Agent Benchmarknot in index	–	–	12.1%#11	–	Vals AI
Legal Research Benchnot in index	–	–	38.9%#32	–	Vals AI
SkillsBenchnot in index	–	–	58.0%#12	–	Vals AI
SREBenchnot in index	–	–	11.1%#13	–	Vals AI
Tax Agent Benchnot in index	–	–	32.1%#19	–	Vals AI
Terminal-Bench 4.0 (Vals)not in index	–	–	19.7%#28	–	Vals AI
Terminal-Bench Sciencenot in index	–	–	8.6%#19	–	Vals AI
Time Horizon Index: KSPnot in index	–	–	18.8%#6	–	Vals AI
Vending-Bench 2not in indexeffort not stated	–	–	–	5093.8#31	Andon Labs
AutomationBenchnot in index	36.5%#101	60.9%#32	59.9%#36	–	Artificial Analysis
GDP.pdf (AA)not in index	18.6%#70	22.8%#44	21.0%#55	–	Artificial Analysis
Harvey LAB-AA v1.1not in index	–	–	0.0%#23	–	Harvey
AA-Briefcase v1.1not in index	–	–	1203#47	–	Artificial Analysis
Terminal-Bench-Science 0.1 (AA)not in index	–	–	10.0%#26	–	Artificial Analysis
CWE-Bench (AA)	–	–	52.5%#11	–	Artificial Analysis
DeepSec-Bench (AA)	–	–	14.5%#14	–	Artificial Analysis
CyberGym end-to-end (AA)	–	–	8.4%#13	–	Artificial Analysis
FACTS Searchnot in indexeffort not stated	–	–	–	76.3%#17	Google

Maths

Benchmark	high	not stated	Source
FrontierMath Tiers 1–3	68.4%#30	–	Epoch AI Benchmarking Hub
FrontierMath Tier 4	21.9%#51	–	Epoch AI Benchmarking Hub
OTIS Mock AIME	98.9%#18	–	Epoch AI Benchmarking Hub
ProofBencheffort not stated	–	48.0%#34	Vals AI
LiveBench Mathematicsnot in index	91.6%#29	–	LiveBench
Arena Maths	1522#4	–	Arena (LMArena)

Knowledge

Benchmark	low	medium	high	not stated	Source
SimpleQA Verified	–	–	69.7%#8	–	Epoch AI Benchmarking Hub
LiveBench 2.0 Data Analysisnot in indexeffort not stated	–	–	–	54.0%#45	LiveBench
LiveBench Researchnot in indexeffort not stated	–	–	–	79.1%#19	LiveBench
LiveBench 1.0 Data Analysisnot in index	–	–	54.0%#64	–	LiveBench
AA-Omniscience	21.4#60	28.6#37	29.6#34	–	Artificial Analysis
MMLU-Pro	–	–	90.2%#5	–	Vals AI
LegalBench	–	–	87.0%#8	–	Vals AI
TaxEval	–	–	74.5%#35	–	Vals AI
PRBench Financeeffort not stated	–	–	–	49.5%#12	Scale AI SEAL
PRBench Legaleffort not stated	–	–	–	48.9%#15	Scale AI SEAL
BioMysteryBenchnot in index	–	–	62.2%#22	–	Vals AI
Excel Modeling Benchmarknot in index	–	–	72.2%#8	–	Vals AI
MedCodenot in index	–	–	48.1%#31	–	Vals AI
MedScribenot in index	–	–	84.5%#34	–	Vals AI
Public Benefits Benchnot in index	–	–	65.3%#21	–	Vals AI
GDP.pdfnot in index	–	23.4%#22	23.2%#23	–	Epoch AI Benchmarking Hub
AA-Omniscience: accuracynot in index	52.2%#52	53.0%#50	54.6%#40	–	Artificial Analysis
AA-Omniscience: non-hallucinationnot in index	35.4%#186	48.1%#129	44.8%#143	–	Artificial Analysis

Instruction following

Benchmark	high	Source
LiveBench Languagenot in index	87.8%#7	LiveBench
LiveBench Instruction Followingnot in index	81.4%#1	LiveBench
Arena Instruction Followingnot in index	1490#14	Arena (LMArena)

Human preference

Benchmark	high	Source
Arena Text	1496#9	Arena (LMArena)
Arena Creative Writingnot in index	1487#7	Arena (LMArena)
Arena Multi-turnnot in index	1503#9	Arena (LMArena)
Arena Longer Queriesnot in index	1510#7	Arena (LMArena)

Multimodal

Benchmark	low	medium	high	not stated	Source
Arena Vision	–	–	1315#7	–	Arena (LMArena)
LiveBench Designnot in indexeffort not stated	–	–	–	72.7%#34	LiveBench
MMMU-Pro	84.5%#19	84.2%#21	85.6%#9	–	Artificial Analysis
MMMU-Pro (Vals)not in index	–	–	89.1%#4	–	Vals AI
MortgageTaxnot in index	–	–	65.3%#43	–	Vals AI
SAGEnot in index	–	–	35.1%#67	–	Vals AI
AudioMCnot in index	–	–	60.4%#1	–	Scale AI SEAL
AudioMC (text output)not in index	–	–	60.4%#1	–	Scale AI SEAL
Blueprint-Bench 2not in indexeffort not stated	–	–	–	38.6%#6	Epoch AI Benchmarking Hub
AA-3Dnot in index	–	–	969#14	–	Artificial Analysis

Long context

Benchmark	low	medium	high	Source	Trend
AA-LCR	80.7%#64	84.0%#11	81.3%#55	Artificial Analysis
MLCRnot in index	–	–	21.7%#25	Artificial Analysis

Composite

Benchmark	low	medium	high	not stated	Source
Epoch Capabilities Indexnot in indexeffort not stated	–	–	–	157.4#20	Epoch AI Benchmarking Hub
LiveBench 2.0effort not stated	–	–	–	67.3%#28	LiveBench
LiveBench 1.0not in index	–	–	75.8%#34	–	LiveBench
AA Intelligence Index v4.3.2	33.5#107	39.8#65	40.9#59	–	Artificial Analysis
Vals Indexnot in index	–	–	56.0#13	–	Vals AI
AA Cyber Indexnot in index	–	–	25.1#14	–	Artificial Analysis

Where it wins

Benchmarks where this configuration ranks in the top five of every configuration measured.

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 Vertex	116 tok/s	3.66 s	$0.825	$4.13	1.0M	–
Google Vertex	102 tok/s	2.75 s	$1.35	$6.75	1.0M	–
Google AI Studio	94 tok/s	2.24 s	$0.750	$3.75	1.0M	–
Google AI Studio	82 tok/s	2.25 s	$0.375	$1.88	1.0M	–
Google Vertex	81 tok/s	2.28 s	$0.750	$3.75	1.0M	–
Google AI Studio	76 tok/s	3.10 s	$1.35	$6.75	1.0M	–
Google Vertex	48 tok/s	13 s	$0.375	$1.88	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 $0.075 per 1M. Reasoning tokens are not modelled.

Workload	Tokens in / out	Cost	With caching	Time
Chat reply	400 / 300	$0.0014	$0.0012	28.3 s
Summarize a 30-page report	12,000 / 600	$0.011	$0.0052	30.4 s
Code edit	6,000 / 1,500	$0.010	$0.0071	36.6 s
Agentic coding session	60,000 / 4,000	$0.060	$0.030	54.0 s
Structured extraction	2,000 / 200	$0.0023	$0.0012	27.6 s