Inspect the GPT-5.6 Compatible Encoding in Python
The GPT-5.6 Sol, Terra, and Luna selections on this page use an o200k_base compatible encoding view. Token IDs describe that vocabulary; they do not establish the full input format of a particular GPT-5.6 API snapshot. Choosing the encoding explicitly makes this local comparison reproducible even when a tokenizer library does not recognize a newer model name.
The example counts the same English and Chinese prompts separately with tiktoken. It adds no chat roles, tool schemas, image inputs, or reasoning markers. Use the result to compare wording and inspect raw token IDs, then confirm complete request usage with OpenAI for the exact model you call.
pythonCompare compatible base-text counts
# pip install tiktoken
import tiktoken
# Compatible base text, not complete GPT-5.6 request usage.
encoding = tiktoken.get_encoding("o200k_base")
prompts = ["Summarize this document.", "请总结这份文档。"]
for prompt in prompts:
token_ids = encoding.encode(prompt)
print(repr(prompt), len(token_ids), token_ids)