DataLearner logo

Gemini 3.8 FlashvsGPT-5.6 Terra

Gemini 3.8 Flash and GPT-5.6 Terra are tied across 4 shared benchmarks: Gemini 3.8 Flash leads on 2, GPT-5.6 Terra leads on 2, with 0 ties and an average score difference of -4.23.

Google Deep Mind
Gemini 3.8 Flash

Google Deep Mind · 2026-09-02 · Multimodal model

OpenAI
GPT-5.6 Terra

OpenAI · 2026-06-26 · Reasoning model

Gemini 3.8 Flash2 wins(50%)(50%)2 winsGPT-5.6 Terra

Benchmark scores

Grouped by capability, sorted by largest gap within each. 4 shared benchmarks.

AI Agent - Tool Usage

Even 2/2
BenchmarkGemini 3.8 FlashGPT-5.6 TerraDiff
Terminal-Bench 4.019.109 / 12Thinking (With Tools)21.527 / 12Max (With Tools)-2.42
Terminal-Bench 2.189.401 / 48Thinking (With Tools)87.408 / 48最高(无工具)+2

Coding and Software Engineer

Gemini 3.8 Flash 1/1
BenchmarkGemini 3.8 FlashGPT-5.6 TerraDiff
DeepSWE73.701 / 33Thinking (With Tools)69.604 / 33极高强度思考(工具)+4.10

Productivity Knowledge

GPT-5.6 Terra 1/1
BenchmarkGemini 3.8 FlashGPT-5.6 TerraDiff
GDPval-AA v21,54513 / 24Thinking (No Tools)1,56611 / 24Max (With Tools)-20.62

Specs

FieldGemini 3.8 FlashGPT-5.6 Terra
PublisherGoogle Deep MindOpenAI
Release date2026-09-022026-06-26
Model typeMultimodal modelReasoning model
ArchitectureDenseDense
ParametersNot availableNot available
Context length1M1.05M
Max output64K128K

API pricing

Prices use DataLearner records when available; missing fields are not inferred.

ItemGemini 3.8 FlashGPT-5.6 Terra
Text input$0.75 / 1M tokens$2 / 1M tokens
Text output$3.75 / 1M tokens$12 / 1M tokens
Cache read$0.075 / 1M tokens$0.2 / 1M tokens
Cache writeNot public$2.5 / 1M tokens

Summary

  • Gemini 3.8 Flashleads in:Coding and Software Engineer (1/1)
  • GPT-5.6 Terraleads in:Productivity Knowledge (1/1)
  • Tied in:AI Agent - Tool Usage

On average across the 4 shared benchmarks, GPT-5.6 Terra scores 4.23 higher.

Largest single-benchmark gap: GDPval-AA v2 — Gemini 3.8 Flash 1,545 vs GPT-5.6 Terra 1,566 (-20.62).

Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.