Updated March 26, 2026· Based on independent benchmark data
GPT-5.2 Codex (xhigh) leads in intelligence with a score of 49.0 vs 45.0. Qwen3.5 397B A17B (Reasoning) is 2.9x cheaper at $0.60/1M tokens vs $1.75/1M. For speed, GPT-5.2 Codex (xhigh) wins at 99 tok/s vs 53 tok/s.
| Metric | GPT-5.2 Codex (xhigh) | Qwen3.5 397B A17B (Reasoning) |
|---|---|---|
| Intelligence Score | 49.0 | 45.0 |
| Coding Score | 43.0 | 41.3 |
| Math Score | N/A | N/A |
| Speed (tok/s) | 99 tok/s | 53 tok/s |
| Latency (TTFT) | 6.86s | 1.46s |
| Input Price / 1M tokens | $1.75 | $0.60 |
| Output Price / 1M tokens | $14 | $3.60 |
| Context Window | N/A |
GPT-5.2 Codex (xhigh) outperforms Qwen3.5 397B A17B (Reasoning) on the intelligence index with a score of 49.0 compared to 45.0. Their coding capabilities are closely matched (43.0 vs 41.3).
GPT-5.2 Codex (xhigh) generates output significantly faster at 99 tok/s compared to Qwen3.5 397B A17B (Reasoning)'s 53 tok/s, making it 1.9x faster for streaming responses. Time to first token is 1.46s for Qwen3.5 397B A17B (Reasoning) vs 6.86s for GPT-5.2 Codex (xhigh), which affects perceived responsiveness in interactive applications.
Qwen3.5 397B A17B (Reasoning) is more affordable at $0.60/1M input tokens ($3.60/1M output), while GPT-5.2 Codex (xhigh) costs $1.75/1M input ($14/1M output). That makes GPT-5.2 Codex (xhigh) 2.9x more expensive per token, which can add up significantly at scale. For a typical workload of 100 requests per day at 2,000 tokens each, GPT-5.2 Codex (xhigh) would cost approximately $10.50/month vs $3.60/month for Qwen3.5 397B A17B (Reasoning) in input costs alone.
Choose GPT-5.2 Codex (xhigh) when you need higher intelligence (49.0), faster output (99 tok/s). Choose Qwen3.5 397B A17B (Reasoning) when you need lower cost.
GPT-5.2 Codex (xhigh) scores higher on coding benchmarks (43.0 vs 41.3), making it the better choice for programming tasks.
Qwen3.5 397B A17B (Reasoning) is cheaper at $0.60/1M input tokens vs $1.75/1M for GPT-5.2 Codex (xhigh).
GPT-5.2 Codex (xhigh) is faster, producing output at 99 tok/s compared to Qwen3.5 397B A17B (Reasoning)'s 53 tok/s.
No, GPT-5.2 Codex (xhigh) does not support image input. Neither model supports image input.
Data last synced: March 26, 2026
| N/A |
| Max Output Tokens | N/A | N/A |
| Input Modalities | Text | Text |
| Output Modalities | Text | Text |
| Free Tier | No | No |
It depends on your priorities. GPT-5.2 Codex (xhigh) scores higher on intelligence (49.0), but Qwen3.5 397B A17B (Reasoning) may be better for specific use cases like budget-conscious projects or speed-critical applications.