Mathematically it comes from the fact that this transformer block is this parallel algorithm. If you batch harder, increase parallelism, you can get higher tokens/s. But you get less throughput. Simultaneously there is also this dial that you can speculatively decode harder with fewer users.
Its true for basically all hardware and most models. You can draw this Pareto curve of how much throughput per GPU vs how many tokens per second per stream. More tokens/s less total throughput.
See this graph for actual numbers:
Token Throughput per GPU vs. Interactivity gpt-oss 120B • FP4 • 1K / 8K • Source: SemiAnalysis InferenceMAX™
I think you skipped the word “total throughout” there right? Cause tok/s is a measure of throughput, so it’s clearer to say you increase throughput/user at the expense of throughput/gpu.
I’m not sure about the comment about speculative decode though. I haven’t served a frontier model but generally speculative decode I believe doesn’t help beyond a few tokens, so I’m not sure you can “speculatively decode harder” with fewer users.
H100 SXM: 3.35 TB/s HBM3
GB200: 8 TB/s HBM3e
2.4x faster memory - which is exactly what they are saying the speedup is. I suspect they are just routing to GB200 (or TPU etc equivalents).
FWIW I did notice _sometimes_ recently Opus was very fast. I put it down to a bug in Claude Code's token counting, but perhaps it was actually just occasionally getting routed to GB200s.
Why does this seem unlikely? I have no doubt they are optimizing all the time, including inference speed, but why could this particular lever not entirely be driven by skipping the queue? It's an easy way to generate more money.
When you add a job with high priority all those chunks will be processed off the queue first by each and every GPU that frees up. It probably leads to more parallelism but... it's the prioritization that led to this happening. It's better to think of this as prioritization of your job leading to the perf improvement.
Here's a good blog for anyone interested which talks about prioritization and job scheduling. It's not quite at the datacenter level but the concepts are the same. Basically everything is thought of as a pipeline. All training jobs are low pri (they take months to complete in any case), customer requests are mid pri and then there's options for high pri. Everything in an AI datacenter is thought of in terms of 'flow'. Are there any bottlenecks? Are the pipelines always full and the expensive hardware always 100% utilized? Are the queues backlogs big enough to ensure full utilization at every stage?
Amazon Bedrock has a similar feature called "priority tier": you get faster responses at 1.75x the price. And they explicitly say in the docs "priority requests receive preferential treatment in the processing queue, moving ahead of standard requests for faster responses".
> codex-5.2 is really amazing but using it from my personal and not work account over the weekend taught me some user empathy lol it’s a bit slow
Let me guess. Quantization?