Compare

o4 vs Gemini 2.5 Ultra

OpenAI vs Google DeepMind. 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.

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

Google DeepMind

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

2M context windowNative videoBest multimodal reasoning

Pricing

Input / 1M$15.00$7.00

Output / 1M$60.00$21.00

Context200K2.0M

Max output100K64K

LicenseProprietaryProprietary

Released2026-02-182026-02-05

Benchmarks

LMSYS ELO—1385.0

MMLU Pro93.890.4

HumanEval92.390.2

SWE-Bench68.058.3

MATH96.292.0

GPQA82.5—

MMMU—82.1

IFEval—89.7

Per-task cost

Summarize a 1-hour meeting transcript$0.255$0.115

Review a 500-line pull request$0.240$0.098

Answer a customer support ticket$0.093$0.038

Extract structured data from a resume$0.135$0.056

Debug a stack trace with context$0.210$0.084

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

Add up to four models, tweak tasks live.