OpenAI GPT-OSS 120B vs Qwen3 235B A22B Instruct 2507 FP8 Throughput
On provider list prices, OpenAI GPT-OSS 120B costs $0.15 per million input tokens against $0.20 for Qwen3 235B A22B Instruct 2507 FP8 Throughput: 1.3x apart. Output is $0.60 against $0.60. On Allocate both bill at list plus the 7% transaction fee.
Specifications and provider list prices from the Allocate catalog, checked 2026-07-08. Billed price is list plus the 7% transaction fee.
What the numbers say
Take 1,000,000 requests a month at 1,200 input and 350 output tokens each. That workload costs $390 a month on OpenAI GPT-OSS 120B and $450 on Qwen3 235B A22B Instruct 2507 FP8 Throughput at list: a gap of $60, or 1.2x.
Qwen3 235B A22B Instruct 2507 FP8 Throughput reads 256K tokens per request against 128K for OpenAI GPT-OSS 120B, 2.0x the window. That decides which one can take whole documents without splitting them.
Choose OpenAI GPT-OSS 120B for
- The lower list price ($0.15 in / $0.60 out per M tokens)
Choose Qwen3 235B A22B Instruct 2507 FP8 Throughput for
- The longer context window (256K vs 128K tokens)
- Fine-tuning under a permissive license (Apache 2.0)
Common questions
Which is cheaper, OpenAI GPT-OSS 120B or Qwen3 235B A22B Instruct 2507 FP8 Throughput?
OpenAI GPT-OSS 120B, on this workload shape. At list prices it is $0.15/$0.60 per million tokens in and out against $0.20/$0.60 for Qwen3 235B A22B Instruct 2507 FP8 Throughput. Billed on Allocate: $0.16/$0.64 against $0.21/$0.64, list plus 7%.
Which has the bigger context window?
Qwen3 235B A22B Instruct 2507 FP8 Throughput: 262,144 tokens (256K) against 131,072 (128K) for OpenAI GPT-OSS 120B.
Can I fine-tune OpenAI GPT-OSS 120B or Qwen3 235B A22B Instruct 2507 FP8 Throughput?
Both publish open weights (OpenAI GPT-OSS 120B: Custom license; Qwen3 235B A22B Instruct 2507 FP8 Throughput: Apache 2.0), so both can be fine-tuned. On Allocate the trained weights stay inside your boundary and belong to you.
Related comparisons
Run the numbers on your workload
Or don’t choose. On Allocate a route name is the contract: point yours at one model today, swap to the other tomorrow, and compare them on your live traffic with per-token metering.