01Overview02Models03Rankings04Compare05Quiz06Guide

Model

GPT-5.5 Pro NO.12

Previous-generation maximum-accuracy OpenAI tier, succeeded by Sol's Ultra mode.

CompareBack to the index

OpenAI · GPT · Reasoning · Closed weights

Specification

Fig.01 — GPT — OpenAI

USA

Context window
1.1Mtokens
Max output
128Ktokens
Input price
$30/ 1M
Output price
$180/ 1M
Throughput
—
Class
Reasoning
Modalities
Text, Image, Pdf
Weights
Closed
Knowledge cutoff
Dec 2025
Released
Apr 2026

Overview

GPT-5.5 Pro is the mandatory-reasoning, maximum-accuracy variant of OpenAI's previous flagship, tuned for the hardest problems where correctness outweighs cost and latency — now succeeded by GPT-5.6 Sol's Pro and Ultra heavy-compute modes. It spends far more compute per query, pushing deeper deliberate reasoning for elite math, science, and analysis. Reserve it for the small slice of tasks that justify its steep pricing; new work should prefer Sol Ultra.

Strengths

  • Top-tier deliberate reasoning accuracy
  • Elite math and science problem solving
  • Frontier coding on hard problems
  • Reliable high-stakes single-shot answers
  • Million-token context with multimodal input

Trade-offs

  • Very expensive at $30/$180 per million tokens
  • High latency from heavy reasoning
  • Overkill for routine work
  • No prompt-cache discount advantage at this tier

Fit

Best for

  • Hardest math and science problems
  • High-stakes analysis where accuracy is paramount
  • Research-grade reasoning
  • Complex algorithmic coding

Not ideal for

  • Cost-sensitive or high-throughput workloads
  • Latency-sensitive interactive apps
  • Routine tasks GPT-5.5 already handles

Capabilities

Ten axes, normalized 0–100 and scored the same way across the whole catalog — so a 78 here means what a 78 means anywhere else on the bench.

Fig.02 — Capability radar

0 — 100

Reasoning
98
Coding
95
Math
97
Writing
91
Knowledge
95
Speed
30
Agentic
94
Vision
89
Multilingual
90
Long Context
92

Benchmarks

Public results, with independent third-party runs marked. Bars normalize percentages against 100 and Elo ratings against a 1500 ceiling.

BenchmarkResult
GPQA DiamondIndependentReasoning · Independent aggregators
94.5%
AIME 2025Math · Vendor-reported
96.5%
LMArena EloIndependentGeneral · LMArena
1510Elo

Alternatives

Models in roughly the same class — the ones worth weighing against this record.

Compare side by side
01GPT-5.6 Sol UltraOpenAISol's maximum-compute mode — cooperating sub-agents for the hardest multi-step work.02GPT-5.5OpenAIPrevious-generation OpenAI flagship — frontier-reliable, now succeeded by GPT-5.6 Sol.03Claude Fable 5AnthropicAnthropic's most capable model — the reliability frontier for the hardest agentic work.

Index

PreviousGPT-5.5NextGPT-5.4
All models/EDUCATION