Gemini 3 Flash Preview

google/gemini-3-flash-preview-20251217

Google: Gemini 3 Flash Preview processes text, image, file, audio, and video inputs and returns text outputs. It has a context length of 1,048,576 tokens and supports tools and reasoning. Input pricing is $0.50 per million tokens, while output pricing is $3.00 per million tokens.

  • tool use
  • structured outputs
  • reasoning

At a glance

providergoogle
releasedDec 17, 2025
context1.0M
input / M$0.50

Specifications

What the API gives you

Context window
1.0M tokens
Max output
66K tokens
Input price
$0.50 / 1M tokens
Output price
$3 / 1M tokens
Cache read
$0.050 / 1M tokens
Modalities
text, image, file, audio, video → text
Knowledge cutoff
—
Released
Dec 17, 2025

Measured evidence

Public benchmark record

Design Arena Coding
1191 Elo2026-10-20
Design Arena Full-stack Agent
1063 Elo2026-10-14
Design Arena Web-app Agent
1110 Elo2026-10-20
Design Arena Website
1196 Elo2026-10-20
FACTS Grounding
59%2026-10-20
FrontierMath Tiers 1-3 (Epoch AI)
51.2% ± 32026-06-11evaluated 2026-06-11
GPQA Diamond (Epoch AI)‡
89.4% ± 2.22026-08-06high effort, evaluated 2026-08-06
GPQA Diamond (Epoch AI)
83.2% ± 1.92025-12-17evaluated 2025-12-17
LMArena Creative Writing
1457 Elo2026-10-15
LMArena Instruction Following
1437 Elo2026-10-15
LMArena Long Query
1452 Elo2026-10-15
LMArena Math
1473 Elo2026-10-15
LMArena Text
1473 Elo2026-10-15
LMArena Text Factuality
1468 Elo2026-10-15
LMArena Vision
1273 Elo2026-10-19
LMArena Vision Diagrams
1293 Elo2026-10-19
LMArena Vision OCR
1285 Elo2026-10-19
LMArena WebDev
1440 Elo2026-10-19
LMArena WebDev Image-to-Web
1450 Elo2026-10-19
OTIS Mock AIME (Epoch AI)‡
95.6% ± 3.12026-08-06high effort, evaluated 2026-08-06
OTIS Mock AIME (Epoch AI)
92.8% ± 3.52025-12-17evaluated 2025-12-17
SimpleQA Verified (Epoch AI)‡
66.8% ± 1.52026-08-27high effort, evaluated 2026-08-27
SWE-bench Verified (verified-500)‡
75.8%2026-10-20mini-SWE-agent 2.0.0, high effort, evaluated 2026-02-17
tau2-bench (1.0.1)‡
27.3%2026-03-02high effort, evaluated 2026-03-02

LMArena data licensed CC-BY-4.0. Aider (Apache-2.0) and SWE-bench (MIT) are open. Design Arena uses blind human preference. ‡ Submitted scaffold plus model result. Compare the harness and effort settings, not only the model name.

Task evidence

How broad is the support?

Each status is exact to this model version and the selected task. Official release claims remain provisional.

  • Best LLM for Frontend & UIEstablished evidence

    73% of intended weighted task evidence across 2 independent sources. 6 of 6 configured evaluation feeds currently have results.

  • Best LLM for Document SummarizationEstablished evidence

    65% of intended weighted task evidence across 2 independent sources. 5 of 6 configured evaluation feeds currently have results.

  • Best LLM for SummarizationEstablished evidence

    65% of intended weighted task evidence across 2 independent sources. 5 of 6 configured evaluation feeds currently have results.

  • Best LLM for Grounded RAGEstablished evidence

    62% of intended weighted task evidence across 3 independent sources. 5 of 7 configured evaluation feeds currently have results.

  • Best LLM for RAG & Long ContextEstablished evidence

    61% of intended weighted task evidence across 2 independent sources. 5 of 6 configured evaluation feeds currently have results.

  • Best LLM for a General AssistantEstablished evidence

    59% of intended weighted task evidence across 2 independent sources. 6 of 6 configured evaluation feeds currently have results.

Show all 38 tasks
  • Best LLM for Competition MathEstablished evidence

    56% of intended weighted task evidence across 3 independent sources. 6 of 6 configured evaluation feeds currently have results.

  • Best LLM for Logic & Problem SolvingEstablished evidence

    55% of intended weighted task evidence across 2 independent sources. 5 of 5 configured evaluation feeds currently have results.

  • Best LLM for Math & ReasoningEstablished evidence

    52% of intended weighted task evidence across 4 independent sources. 7 of 7 configured evaluation feeds currently have results.

  • Best LLM for CodingEstablished evidence

    42% of intended weighted task evidence across 3 independent sources. 7 of 7 configured evaluation feeds currently have results.

  • Best LLM for Debugging & Code ReviewEstablished evidence

    28% of intended weighted task evidence across 2 independent sources. 5 of 5 configured evaluation feeds currently have results.

  • Best LLM for Business WritingProvisional evidence

    67% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from LiveBench Instruction Following. 4 of 4 configured evaluation feeds currently have results.

  • Best LLM for Marketing CopyProvisional evidence

    67% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from LiveBench Instruction Following. 4 of 4 configured evaluation feeds currently have results.

  • Best LLM for Chat & RoleplayProvisional evidence

    67% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from LiveBench Instruction Following or LiveBench Language. 4 of 4 configured evaluation feeds currently have results.

  • Best LLM for Roleplay & CharacterProvisional evidence

    67% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from LiveBench Instruction Following or LiveBench Language. 4 of 4 configured evaluation feeds currently have results.

  • Best LLM for OCR & DocumentsProvisional evidence

    65% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from Structured-output eval or VLMEvalKit tasks. 3 of 5 configured evaluation feeds currently have results.

  • Best LLM for Charts & DiagramsProvisional evidence

    65% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from Structured-output eval or VLMEvalKit tasks. 3 of 5 configured evaluation feeds currently have results.

  • Best LLM for Vision & DocumentsProvisional evidence

    60% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from Structured-output eval or VLMEvalKit tasks. 4 of 6 configured evaluation feeds currently have results.

  • Best LLM for Document ParsingProvisional evidence

    58% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from Structured-output eval or VLMEvalKit tasks. 3 of 5 configured evaluation feeds currently have results.

  • Best LLM for Creative WritingProvisional evidence

    56% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from LiveBench Instruction Following or LiveBench Language. 4 of 4 configured evaluation feeds currently have results.

  • Best LLM for WritingProvisional evidence

    56% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from LiveBench Instruction Following or LiveBench Language. 4 of 4 configured evaluation feeds currently have results.

  • Best LLM for Image UnderstandingProvisional evidence

    50% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from LiveBench Reasoning or VLMEvalKit tasks. 2 of 3 configured evaluation feeds currently have results.

  • Best LLM for Massive ContextProvisional evidence

    47% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from LongBench v2. 2 of 3 configured evaluation feeds currently have results.

  • Best LLM for Code CompletionProvisional evidence

    36% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from Aider Polyglot or LiveBench Coding. 3 of 3 configured evaluation feeds currently have results.

  • Best Open-Weight LLM for Self-HostingProvisional evidence

    27% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from LiveBench Instruction Following or Route reliability. 2 of 3 configured evaluation feeds currently have results.

  • Best Value LLMProvisional evidence

    25% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from LiveBench Instruction Following or Route reliability. 2 of 3 configured evaluation feeds currently have results.

  • Best Local LLM for CodingProvisional evidence

    21% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from Aider Polyglot or LiveBench Coding. 4 of 4 configured evaluation feeds currently have results.

  • Best LLM for Tool & Function CallingProvisional evidence

    21% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from Berkeley Function Calling or LMArena Agent. 6 of 7 configured evaluation feeds currently have results.

  • Best LLM for Agentic CodingProvisional evidence

    20% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from LiveBench Agentic Coding or LMArena Agent. 7 of 7 configured evaluation feeds currently have results.

  • Best LLM for Fast SummariesProvisional evidence

    17% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from LiveBench Instruction Following or Route reliability. 2 of 3 configured evaluation feeds currently have results.

  • Best LLM for Literary TranslationProvisional evidence

    17% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from LiveBench Instruction Following or LiveBench Language. 3 of 4 configured evaluation feeds currently have results.

  • Best LLM for Everyday TranslationProvisional evidence

    17% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from LiveBench Instruction Following or LiveBench Language. 3 of 4 configured evaluation feeds currently have results.

  • Best LLM for TranslationProvisional evidence

    17% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from LiveBench Instruction Following or LiveBench Language. 3 of 4 configured evaluation feeds currently have results.

  • Best LLM for JSON & Schema OutputProvisional evidence

    12% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from Berkeley Function Calling or LiveBench Data Analysis. 3 of 4 configured evaluation feeds currently have results.

  • Best LLM for Structured Data ExtractionProvisional evidence

    12% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from Berkeley Function Calling or LiveBench Data Analysis. 3 of 4 configured evaluation feeds currently have results.

  • Best Cheap LLM for High VolumeProvisional evidence

    10% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from LiveBench Instruction Following or Route reliability. 2 of 3 configured evaluation feeds currently have results.

  • Cheapest Capable LLMProvisional evidence

    10% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from LiveBench Instruction Following or Route reliability. 2 of 3 configured evaluation feeds currently have results.

  • Best LLM for AI AgentsProvisional evidence

    9% of intended weighted task evidence across 1 independent source; broader independent confirmation is still missing from Berkeley Function Calling or LiveBench Agentic Coding. 9 of 9 configured evaluation feeds currently have results.

Recommendation coverage

Where this model appears

Not currently included in a published recommendation. The task evidence panel above says exactly what is confirmed and what is still missing.

Catalog data via OpenRouter and models.dev. Last checked within the past 6 hours.