Atlas

Models

← All models

GPT-5.1-Codex-Max

OpenAI · 2025-11-18

GPT-5.1-Codex-Max is OpenAI's frontier agentic coding model, optimized for long-running tasks. It was OpenAI's first model natively trained to operate across multiple context windows through compaction, and it supports variable reasoning effort including Extra High (xhigh).

Benchmark scores

BenchmarkScore
ALE-Bench1538
App-Bench38.4
BuseyBench SVG4.9
Capability151.5
Epoch Capabilities Index (ECI)150
IOI21.4
IOI - IOI 202414.5
IOI - IOI 202528.3
LiveCodeBench (Vals AI public dataset)83.6
LiveCodeBench (Vals Public) - Easy97.5
LiveCodeBench (Vals Public) - Hard65.4
LiveCodeBench (Vals Public) - Medium87.7
Market-Bench4243
Market-Bench - Dynamic Delta Hedging (Best of 5)1371
Market-Bench - Dynamic Delta Hedging (Mean of 5)10496
Market-Bench - Pairs Mean-Reversion (Best of 5)89.4
Market-Bench - Pairs Mean-Reversion (Mean of 5)137
Market-Bench - Single-Stock Scheduled Execution (Best of 5)0.0
Market-Bench - Single-Stock Scheduled Execution (Mean of 5)845
METR Time Horizons162
PostTrainBench Average19.7
ProofBench9.0
SnakeBench Average Apples12.3
SnakeBench Best Apples30.0
SnakeBench Rating36.4
SnakeBench Total Apples392
SnakeBench Win Rate96.9
Terminal-Bench 2.060.4
Vibe Code Bench v1.122.2
Loading Atlas data…