GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 827 -5.3%
QWEN-38X 824 —
CL-OP55X 820 —
GPT-6A 820 —
GROK-46H 820 -5.2%
GLM-5 784 -8.4%
KIMI-K3X 742 -8.4%
CL-FAB5H 742 -5.7%
CL-OP5H 718 -6%
CL-OP5X 708 -18.2%
CL-OP46H 696 -6.2%
CL-OP47H 688 -6.1%
GEM-38FH 677 +0.1%
GEM-37FH 655 -24.3%
GPT-56S 619 —
GPT-55H 580 —
CL-OP47 579 -0.7%
INKL 531 —
GEM-31P 512 —
GEM-3P 498 —
CL-OP46 496 —
CL-OP48 489 -0.2%
GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 827 -5.3%
QWEN-38X 824 —
CL-OP55X 820 —
GPT-6A 820 —
GROK-46H 820 -5.2%
GLM-5 784 -8.4%
KIMI-K3X 742 -8.4%
CL-FAB5H 742 -5.7%
CL-OP5H 718 -6%
CL-OP5X 708 -18.2%
CL-OP46H 696 -6.2%
CL-OP47H 688 -6.1%
GEM-38FH 677 +0.1%
GEM-37FH 655 -24.3%
GPT-56S 619 —
GPT-55H 580 —
CL-OP47 579 -0.7%
INKL 531 —
GEM-31P 512 —
GEM-3P 498 —
CL-OP46 496 —
CL-OP48 489 -0.2%
← Back to feed

GPT-5.5 Costs 49-92% More in Practice — OpenRouter's Switcher Cohort Study Breaks Down the Real Price

OpenAI’s sticker price for GPT-5.5 is double GPT-5.4: $5/M input and $30/M output, up from $2.50 and $15. The company argued the model is less verbose, keeping net costs closer to flat. OpenRouter ran the numbers on users who actually switched and found the reality is messier.

The Methodology

OpenRouter identified a “switcher cohort” — users whose top model by request count was GPT-5.4 before the launch, who then moved to GPT-5.5 as their primary model. They tracked the same users across the same workflows before and after, normalising to cost per million OpenRouter tokens to control for prompt length variation. GPT-5.4 and 5.5 share a tokenizer family, so no tokenizer correction was needed.

Verbosity: Only Cuts Long-Context Prompts

GPT-5.5 does produce shorter completions — but only when the input is long:

Prompt SizeGPT-5.4 Median CompletionGPT-5.5 Median CompletionChange
< 2K tokens121129+7%
2K – 10K140213+52%
10K – 25K211143-32%
25K – 50K185150-19%
50K – 128K188136-28%
128K+215143-34%

Under 10K tokens, the model is actually more verbose than its predecessor. The efficiency gains OpenAI cited apply to the long-context use case only. Short prompt users, who make up the majority of API calls by count, see no offset.

Actual Cost Impact

Prompt SizeGPT-5.4 Avg $/MGPT-5.5 Avg $/MIncrease
< 2K tokens$4.89$9.37+92%
2K – 10K$2.25$3.81+69%
10K – 25K$1.42$2.15+51%
25K – 50K$1.02$1.65+62%
50K – 128K$0.74$1.10+49%
128K+$0.71$1.31+85%

The cheapest case is 50K-128K tokens at +49%. Short-context developers see costs nearly double. Heavy-context developers — the segment most likely to benefit from reduced verbosity — still absorb 49-62% increases.

Context

OpenRouter ran the same analysis on Opus 4.7 earlier. That study found a 27% net cost increase over Opus 4.6 in the agentic sweet spot. GPT-5.5 lands at a steeper premium in every bucket. For cost-sensitive deployments that can tolerate slightly lower capability, GPT-5.4 and DeepSeek V4 remain the incumbents. For teams that have already moved to 5.5, the data suggests auditing prompt lengths first — the savings narrative only holds for long-context heavy users.

OpenAI noted separately that GPT-5.5 scores 82.7% on Terminal-Bench 2.0 and 88.7% on SWE-Bench Verified, ahead of every other model. Whether the performance delta justifies 49-92% higher operating costs depends entirely on the task.