BenchLeader
Menu
LeaderboardsModelsProvidersBenchmarksCompareReleasesBlogMethodology
GoogleReasoning model

Gemini 2.5 Pro

Best configuration ranks #205 of 610 on the BenchLeader Index at 54.1 ±3.4. Last measured 8 Sept 2026. Released 25 Mar 2025.

Blended price
$3.44/M
$1.25 in · $10.00 out
Output speed
123 tok/s
First answer
23 s
first token 2.26 s
Context
1.0M
How it scores by categoryDashed line = average model (50). One step of 15 = one standard deviation.
  1. Overall index54
  2. Reasoning54
  3. Coding49
  4. Agents & tools41
  5. Maths51
  6. Knowledge54
  7. Instruction following51
  8. Human preference64
  9. Multimodal64
  10. Long context67
  11. Composite50

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
thinking––123 tok/s23 s$0.0035Reasoning 38
defaultbest54.1#205123 tok/s23 s$0.0035Agents & tools 41 · Coding 49 · Composite 50 · Human preference 64 · Instruction following 51 · Knowledge 54 · Long context 67 · Maths 51 · Multimodal 64 · Reasoning 54

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

BenchmarkthinkingdefaultSourceTrend
GPQA Diamond–85.3%#79Epoch AI Benchmarking Hub
Humanity's Last Exam–21.6%#16Scale AI / CAIS
SimpleBench–62.4%#21SimpleBench
LMArena Hard Prompts–1458#84LMArena
GPQA Diamond (AA)not in index–84.4%#141Artificial Analysis
Humanity's Last Exam (AA)not in index–22.5%#147Artificial Analysis
GPQA Diamond (Vals)not in index–80.8%#64Vals AI
Kagi LLM Benchmark–70.3%#26Kagi LLM Benchmark
ARC-AGI-141.0%#13033.0%#143ARC Prize
ARC-AGI-24.9%#1203.8%#131ARC Prize

Coding

BenchmarkthinkingdefaultSourceTrend
SWE-bench Verified (Epoch)–57.6%#31Epoch AI Benchmarking Hub
SciCode–42.8%#95SciCode
WeirdML–54.0%#59WeirdML
GSO-Bench–3.9%#25GSO-Bench
LMArena Coding–1465#105LMArena
LMArena WebDev–1226#112LMArena
SciCode (AA)not in index–46.3%#86Artificial Analysis
LiveCodeBench–79.2%#71Vals AI
IOI–17.1%#27Vals AI
SWE-bench (Vals)not in index–54.4%#78Vals AI
Aider Polyglot–83.1%#3Aider polyglot leaderboard
SWE-bench Verified (bash only)–53.6%#31SWE-bench
SWE-bench Verified (any scaffold)not in index–53.6%#39SWE-bench

Agents & tools

BenchmarkthinkingdefaultSourceTrend
Terminal-Bench–32.6%#44Terminal-Bench
GDPval–23.3%#9OpenAI
Remote Labor Index–0.8%#12Scale AI / CAIS
APEX-Agents–6.6%#55Mercor
Terminal-Bench Hard–26.5%#126Artificial Analysis
τ²-Bench Telecom (AA)not in index–54.1%#182Artificial Analysis

Maths

BenchmarkthinkingdefaultSourceTrend
FrontierMath Tiers 1–3–24.6%#66Epoch AI Benchmarking Hub
FrontierMath Tier 4–0.0%#57Epoch AI Benchmarking Hub
OTIS Mock AIME–84.7%#82Epoch AI Benchmarking Hub
MATH Level 5–95.9%#12Epoch AI Benchmarking Hub
AIME (Vals)–85.8%#35Vals AI
IMO 2025–31.6%#2MathArena
MathArena Apex–0.5%#41MathArena

Knowledge

BenchmarkthinkingdefaultSourceTrend
AA-Omniscience–-16.4#166Artificial Analysis
MMLU-Pro–84.1%#65Vals AI
LegalBench–84.3%#31Vals AI
CorpFin–60.8%#63Vals AI
TaxEval–72.9%#56Vals AI
MedQA–93.1%#25Vals AI
PRBench Finance–38.9%#26Scale AI SEAL
PRBench Legal–41.4%#19Scale AI SEAL
MultiNRC–45.1%#15Scale AI SEAL

Instruction following

BenchmarkthinkingdefaultSourceTrend
IFBench–48.7%#179Artificial Analysis
MultiChallenge–53.6%#20Scale AI SEAL
TutorBench–55.6%#3Scale AI SEAL

Human preference

BenchmarkthinkingdefaultSourceTrend
LMArena Text–1446#72LMArena

Multimodal

BenchmarkthinkingdefaultSourceTrend
LMArena Vision–1262#41LMArena
MMMU-Pro–74.9%#79Artificial Analysis
VISTA–54.6%#1Scale AI SEAL
MMMU (validation)–79.6%#7MMMU
MMMU-Pro (official)not in index–68.0%#12MMMU

Long context

BenchmarkthinkingdefaultSourceTrend
Fiction.LiveBench 120k–90.6%#4Fiction.live
AA-LCR–69.0%#176Artificial Analysis

Composite

BenchmarkthinkingdefaultSourceTrend
Epoch Capabilities Indexnot in index–145.3#70Epoch AI Benchmarking Hub
AA Intelligence Index–16.7#207Artificial 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.

ProviderSpeedFirst tokenInput $/MOutput $/MContextQuantisation
Google Vertex (Global) (ZDR)87 tok/s2.32 s$1.25$10.001.0M–
Google Vertex (Global) (ZDR)87 tok/s2.19 s$2.25$18.001.0M–

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 $0.125 per 1M. Reasoning tokens are not modelled.

WorkloadTokens in / outCostWith cachingTime
Chat reply400 / 300$0.0035$0.003225.1 s
Summarise a 30-page report12,000 / 600$0.021$0.01127.5 s
Code edit6,000 / 1,500$0.022$0.01734.8 s
Agentic coding session60,000 / 4,000$0.115$0.06455.2 s
Structured extraction2,000 / 200$0.0045$0.002824.3 s

See also

Cheapest 1M contextBest for codingBest for agentsAll Google models

Data as of 9 Sept 2026. Compare these configurations.