Summary
The max_input_tokens for gpt-5-pro is set to 128,000 in model_list.json, but OpenAI's official documentation states a 400,000-token context window. The current value appears to have been computed as 400,000 − 272,000 (context window − max_output_tokens), but the catalog convention is max_input_tokens = context_window.
This is the same root cause as the gpt-5/gpt-5-mini/gpt-5-nano discrepancy (see issue #915) — those models have 272,000 instead of 400,000. gpt-5-pro is filed separately because it exceeds the 3-model-per-issue limit.
Affected Model
| Model ID |
Current max_input_tokens |
Correct max_input_tokens |
max_output_tokens |
Line |
gpt-5-pro |
128,000 |
400,000 |
272,000 (correct) |
~618 |
All other fields (format, flavor, pricing, max_output_tokens, providers) are correct and unchanged.
Convention Evidence
| Model |
max_input_tokens |
max_output_tokens |
Sum |
Convention |
gpt-5.5 |
1,050,000 |
128,000 |
1,178,000 |
✅ max_input = context_window |
gpt-5.4-mini |
400,000 |
128,000 |
528,000 |
✅ max_input = context_window |
claude-sonnet-5 |
1,000,000 |
128,000 |
1,128,000 |
✅ max_input = context_window |
gpt-5-pro |
128,000 |
272,000 |
400,000 |
❌ = context_window − max_output |
The catalog stores independent maximums for input and output; their sum is allowed to exceed the context window. The implicit constraint (input + output ≤ context_window) is not enforced through the max_input_tokens field.
Verification Checklist
| Check |
Status |
Detail |
| Cross-source |
✅ |
Context window confirmed on gpt-5-pro model page — "400,000 context window", "272,000 max output tokens" |
| Already fixed |
✅ |
Entry currently shows max_input_tokens: 128000 in model_list.json |
| ID format |
✅ |
Existing entry — no ID change |
| Duplicate check |
✅ |
No open issue covers gpt-5-pro max_input_tokens. Issue #915 covers the sibling models (gpt-5, gpt-5-mini, gpt-5-nano) |
Local Files Inspected
packages/proxy/schema/model_list.json — confirmed current value at line ~618packages/proxy/scripts/sync_models.ts — gpt-5-pro is not in SYNC_PRESERVED_FIELDS
{
"kind": "stale_metadata",
"provider": "openai",
"models": ["gpt-5-pro"],
"status": "active",
"model_specs": {
"gpt-5-pro": {
"format": "openai",
"flavor": "chat",
"multimodal": true,
"input_cost_per_mil_tokens": 15,
"output_cost_per_mil_tokens": 120,
"displayName": "GPT-5 Pro",
"reasoning": true,
"max_input_tokens": 400000,
"max_output_tokens": 272000,
"available_providers": ["openai", "azure"]
}
},
"source_urls": [
"https://developers.openai.com/api/docs/models/gpt-5-pro"
]
}
Summary
The
max_input_tokensforgpt-5-prois set to 128,000 inmodel_list.json, but OpenAI's official documentation states a 400,000-token context window. The current value appears to have been computed as400,000 − 272,000(context window − max_output_tokens), but the catalog convention ismax_input_tokens = context_window.This is the same root cause as the
gpt-5/gpt-5-mini/gpt-5-nanodiscrepancy (see issue #915) — those models have 272,000 instead of 400,000.gpt-5-prois filed separately because it exceeds the 3-model-per-issue limit.Affected Model
gpt-5-proAll other fields (format, flavor, pricing, max_output_tokens, providers) are correct and unchanged.
Convention Evidence
gpt-5.5gpt-5.4-miniclaude-sonnet-5gpt-5-proThe catalog stores independent maximums for input and output; their sum is allowed to exceed the context window. The implicit constraint (input + output ≤ context_window) is not enforced through the
max_input_tokensfield.Verification Checklist
model_list.jsonLocal Files Inspected
packages/proxy/schema/model_list.json— confirmed current value at line ~618packages/proxy/scripts/sync_models.ts—gpt-5-prois not inSYNC_PRESERVED_FIELDS{ "kind": "stale_metadata", "provider": "openai", "models": ["gpt-5-pro"], "status": "active", "model_specs": { "gpt-5-pro": { "format": "openai", "flavor": "chat", "multimodal": true, "input_cost_per_mil_tokens": 15, "output_cost_per_mil_tokens": 120, "displayName": "GPT-5 Pro", "reasoning": true, "max_input_tokens": 400000, "max_output_tokens": 272000, "available_providers": ["openai", "azure"] } }, "source_urls": [ "https://developers.openai.com/api/docs/models/gpt-5-pro" ] }