BenchLeader
Menu
LeaderboardsModelsProvidersBenchmarksCompareReleasesBlogMethodology
Google

Gemini 2.0 Flash

Best configuration ranks #398 of 610 on the BenchLeader Index at 45.1 ±5.0 (thinking reasoning effort). Last measured 1 Sept 2026. Released 5 Feb 2025.

Blended price
$0.175/M
$0.100 in · $0.400 out
Output speed
–
First answer
–
Context
2M
How it scores by categoryDashed line = average model (50). One step of 15 = one standard deviation.
  1. Overall index45
  2. Reasoning38
  3. Coding32
  4. Maths49
  5. Knowledge51
  6. Multimodal56
  7. Long context37
  8. Composite41

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. “Default” means the publisher did not say which setting was used.

EffortIndexRankSpeedFirst answerChat reply costCategories
thinkingbest45.1#398––$0.0002Coding 32 · Composite 41 · Knowledge 51 · Long context 37 · Maths 49 · Multimodal 56 · Reasoning 38
default42.3#462––$0.0002Agents & tools 37 · Coding 31 · Composite 40 · Human preference 53 · Instruction following 32 · Knowledge 41 · Long context 48 · Maths 46 · Multimodal 49 · Reasoning 38

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

BenchmarkthinkingdefaultSource
GPQA Diamond57.1%#18364.1%#170Epoch AI Benchmarking Hub
Humanity's Last Exam6.6%#35–Scale AI / CAIS
SimpleBench30.7%#6718.9%#80SimpleBench
LMArena Hard Prompts–1360#184LMArena
GPQA Diamond (AA)not in index70.1%#28463.6%#335Artificial Analysis
Humanity's Last Exam (AA)not in index6.3%#3224.3%#434Artificial Analysis
GPQA Diamond (Vals)not in index–65.2%#99Vals AI
ARC-AGI-2–1.3%#153ARC Prize

Coding

BenchmarkthinkingdefaultSource
WeirdML–25.8%#129WeirdML
LMArena Coding–1364#204LMArena
LiveCodeBench–43.6%#117Vals AI
Aider Polyglot18.2%#3722.2%#35Aider polyglot leaderboard
SWE-bench Verified (bash only)–13.5%#42SWE-bench
SWE-bench Verified (any scaffold)not in index–52.2%#41SWE-bench

Agents & tools

BenchmarkthinkingdefaultSource
Terminal-Bench Hard–3.8%#283Artificial Analysis
τ²-Bench Telecom (AA)not in index–29.5%#254Artificial Analysis

Maths

BenchmarkthinkingdefaultSource
OTIS Mock AIME57.8%#15031.1%#187Epoch AI Benchmarking Hub
MATH Level 5–82.2%#31Epoch AI Benchmarking Hub
AIME (Vals)–29.8%#71Vals AI
MGSM–89.0%#48Vals AI

Knowledge

BenchmarkthinkingdefaultSource
AA-Omniscience–-42.5#268Artificial Analysis
MMLU-Pro–77.4%#100Vals AI
LegalBench–78.4%#94Vals AI
CorpFin–33.7%#124Vals AI
TaxEval69.8%#9367.7%#102Vals AI
MedQA–81.5%#67Vals AI

Instruction following

BenchmarkthinkingdefaultSource
IFBench–40.2%#251Artificial Analysis
MultiChallenge–36.4%#29Scale AI SEAL

Human preference

BenchmarkthinkingdefaultSource
LMArena Text–1360#169LMArena

Multimodal

BenchmarkthinkingdefaultSource
LMArena Vision–1158#95LMArena
VISTA45.5%#2340.0%#39Scale AI SEAL
MMMU (validation)–71.7%#20MMMU

Long context

BenchmarkthinkingdefaultSource
Fiction.LiveBench 120k37.5%#2662.5%#13Fiction.live
AA-LCR–31.3%#319Artificial Analysis

Composite

BenchmarkthinkingdefaultSource
Epoch Capabilities Indexnot in index135.4#117134.7#120Epoch AI Benchmarking Hub
AA Intelligence Index9.4#3298.9#344Artificial Analysis

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. Reasoning tokens are not modelled.

WorkloadTokens in / outCostWith cachingTime
Chat reply400 / 300$0.0002––
Summarise a 30-page report12,000 / 600$0.0014––
Code edit6,000 / 1,500$0.0012––
Agentic coding session60,000 / 4,000$0.0076––
Structured extraction2,000 / 200$0.0003––

See also

Best under $2/MBest under $0.50/MCheapest 1M contextAll Google models

Data as of 9 Sept 2026. Compare these configurations.