Compare

Gemini 2.5 Ultra vs o4

Google DeepMind vs OpenAI. Specs, benchmarks, and real per-task cost — all in one page.

Verdict

Gemini 2.5 Ultra is ~2.5× cheaper on a 3:1 input:output blend.

Google DeepMind

2M-token context, native video understanding, and Google’s deepest multimodal stack. The long-context king.

2M context windowNative videoBest multimodal reasoning

OpenAI

Deep-reasoning model. Spends more tokens thinking to crush math, science, and hard coding problems.

State-of-the-art math and scienceExtremely high-ceiling reasoning

Pricing

Input / 1M$7.00$15.00

Output / 1M$21.00$60.00

Context2.0M200K

Max output64K100K

LicenseProprietaryProprietary

Released2026-02-052026-02-18

Benchmarks

LMSYS ELO1385.0—

MMLU Pro90.493.8

HumanEval90.292.3

SWE-Bench58.368.0

MATH92.096.2

GPQA—82.5

MMMU82.1—

IFEval89.7—

Per-task cost

Summarize a 1-hour meeting transcript$0.115$0.255

Review a 500-line pull request$0.098$0.240

Answer a customer support ticket$0.038$0.093

Extract structured data from a resume$0.056$0.135

Debug a stack trace with context$0.084$0.210

Per-call cost using published token counts for each task. Real-world prompts vary.

Add up to four models, tweak tasks live.