Models/GPT-5.3 Codex Spark
OpenAI/GPT-5/Feb 5, 2026

GPT-5.3 Codex Spark

Speed-optimized distillation of GPT-5.3 Codex running on Cerebras WSE-3 hardware. Achieves 1,000+ tokens/sec — 15x faster than the full Codex — while matching its benchmark scores on SWE-Bench Pro (56.8%) and Terminal-Bench 2.0 (77.3%). 50% faster TTFT and 80% reduction in roundtrip overhead.

textcodereasoningtool-use

Arena ELO

Input Price

Output Price

Speed

1000 t/s

Context

128K

Latency

Capability Assessment

SWE-Bench Pro80.0%
Terminal-Bench 2.077.3%

Comparative Analysis

MetricGPT-5.3 Codex SparkClaude Opus 4.6Gemini 3.1 ProGrok 4
SWE-bench80.0%80.8%80.6%72.0%
AIME 202594.0%100.0%91.2%94.0%
GPQA Diamond91.3%94.3%88.0%
MMLU91.0%92.6%
Input $/M$5$2$3
Output $/M$25$12$15

Source Material