OpenAImodel systemcard

GPT-5.6 Terra

#8of 27 · general

effort ladder collapsed

1.1Mcontext

Proprietarynon-commercial

2026-07-09released

2026-02-16knowledge cutoff

Standing

dot = this model · shaded = the field's middle 60% · track = full field

General

59.2#8/27

Math

67.7#7/27

Coding

61.2#9/27

Health

31.1#18/26

Engineering

50.0#9/17

Legal

49.5#13/26

Instruct

28.9#20/27

Agents

59.0#10/27

Professional

39.2#16/27

Multimodal

40.3#11/19

Hallucination

18.2#22/27

Speed

84.3#3/22

Value

50.0#11/27

Measured

≈ imputed · amber dot = lab-self-reported among the sources

GPQA Diamond92.7%#8

HLE42.9%#10

MMLU-Pro86.7%#16

LiveBench77.9%#7

SWE-bench V75.2%#23

LiveCodeBench85.9%#12

SciCode53.9%#9

CritPt30.0%#2

Terminal-Bench80.7%#5

APEX Agents38.9%#6

GDPval53.9%#9

AA Index56.6#7

Vals Index65.1#10

Epoch Capabilities Index159.0#5

Operations

$2 / $12$ per 1M tokens, in / out

$4.5blended

133tokens / sec

$0.35$ per solved task

~1484arena elo · predicted, never voted on

Series

OpenAI's models on the board, by release

GPT-5.52026-04-23 · 67GPT-5.6 Sol2026-07-09 · 85GPT-5.6 Terra2026-07-09 · 59GPT-5.6 Luna2026-07-09 · 50

Neighbors

nearest capability fingerprints · deltas describe the neighbor

Grok 4.5→ follows instructions better · cheaper Claude Sonnet 5→ stronger agent GPT-5.6 Luna→ weaker math · cheaper GLM-5.2→ more factual · cheaper Qwen3.8 Max→ weaker math · cheaper