Back to the index
Google DeepMind · Gemini
No. 12

Gemini 3.5 Pro

Google's newest flagship — a 2M-token context and a Deep Think reasoning mode.

FrontierClosed weightsText, Image, Audio, Video, Pdf
Field notes
Context window
2Mtokens
Max output
66Ktokens
Input price
$15/ 1M tokens
Output price
$60/ 1M tokens
Modalities
Text, Image, Audio, Video, Pdf
Knowledge cutoff
Jan 2026
Released
Jul 2026
Approx. speed
Overview

What this model is

Gemini 3.5 Pro is Google DeepMind's newest flagship, unveiled at Google I/O 2026 and reaching general availability in July 2026 with the largest production context window (2M tokens) and a gated 'Deep Think' extended-reasoning mode. It is positioned as the multimodal and ultra-long-context leader. As a brand-new release it is effectively pre-benchmark — Google had not published official evals or final pricing at launch — so the figures here are provisional, and its real-world long-horizon agentic-coding reliability (historically Gemini's weak spot versus the Claude/GPT frontier) remains to be proven independently.

Strengths

  • Largest production context window (2M tokens)
  • Deep Think extended-reasoning mode
  • Natively multimodal (image/audio/video/PDF)
  • Frontier knowledge, math, and vision pedigree

Trade-offs

  • Pricing and benchmarks unconfirmed at launch (provisional)
  • Deep Think gated to top/enterprise tiers
  • Gemini agentic-coding reliability historically below Claude/GPT
  • Limited-preview rollout at GA

Best for

  • Ultra-long-context document and codebase analysis
  • Multimodal reasoning over image, audio, and video
  • Deep research and math/science

Not ideal for

  • Unsupervised long-horizon coding agents (unproven)
  • Cost-sensitive high-volume workloads
Capabilities

Capability profile

Normalized 0–100 scores, comparable across the whole catalog.

Reasoning92Coding80Math95Writing90Knowledge95Speed55Agentic78Vision96Multilingual92Long Context98
Reasoning
92
Coding
80
Math
95
Writing
90
Knowledge
95
Speed
55
Agentic
78
Vision
96
Multilingual
92
Long Context
98
Benchmarks

How it scores

Public benchmark results, with independent third-party results where available. Bars normalize percentages to 100 and Elo ratings to a 1500 ceiling.

No public benchmark scores recorded for this model.

Alternatives

Comparable models

Models in a roughly similar class — worth weighing against this one.