This page puts price and speed for five models side by side, all called through one OpenAI-compatible base URL, https://api.tokenlab.sh/v1, with one API key. We are an AI team that runs TokenLab, so check the numbers against the live pages. Prices are USD per 1 million tokens, read from the TokenLab pricing API (registry revision 2026-11-04) and, for OpenRouter, from the per-endpoint pricing in its public API on 2026-11-04.
Price: TokenLab Verified vs list price
| Model | TokenLab Verified (in / out) | Vendor list price (in / out) | OpenRouter standard endpoint (in / out) |
|---|---|---|---|
| gpt-6.1-terra | $0.30 / $1.50 | $1.00 / $5.00 | $1.00 / $5.00 |
| claude-sonnet-5-5 | $0.70 / $3.50 | $2.00 / $10.00 | $2.00 / $10.00 |
| grok-4.7 | $1.00 / $3.00 | $2.00 / $6.00 | $2.00 / $6.00 |
| gemini-3.8-flash | $0.375 / $1.875 | $0.75 / $3.75 | $0.75 / $3.75 |
TokenLab Verified is 30% of list price for gpt-6.1-terra, 35% for claude-sonnet-5-5 and 50% for Grok and Gemini. Tiers apply by prompt size: gpt-6.1-terra has a higher rate above 272K input tokens and grok-4.7 above 200K; the figures here are the standard tier. The OpenRouter column is the standard (non-flex, non-priority, non-batch) endpoint price. OpenRouter also lists cheaper flex endpoints for some models, for example gemini-3.8-flash at $0.375 / $1.875 and gpt-6.1-terra at $0.50 / $2.50, which are slower or best-effort tiers; compare like with like. OpenRouter says it passes provider prices through; it also charges a fee when you buy credits, which we do not quote. Verified is TokenLab's lower-priced supply; Official is priced at the maker's list price. See the delivery tiers guide. Not every model has a Verified price: deepseek-v4.1-flash, used in the speed test below, is billed at the maker's list price on TokenLab, so it is not in the price table.
Speed: time to first token
We called each model 20 times with streaming on, one request per model in turn, between 16:38 and 16:49 Beijing time on 2026-11-04. Same prompt each time: a short English text with a request for about 120 words. max_tokens was 800. The client ran in Los Angeles and called api.tokenlab.sh. Times are in milliseconds.
| Model | First token p50 / p95 | First answer text p50 / p95 | Finished p50 / p95 | Runs with answer text |
|---|---|---|---|---|
| deepseek-v4.1-flash | 1174 / 1768 | 3049 / 5958 | 3976 / 6583 | 19 / 20 |
| claude-sonnet-5-5 | 1555 / 1915 | 1555 / 1915 | 4234 / 4540 | 19 / 20 |
| grok-4.7 | 2110 / 2846 | 5044 / 11125 | 7999 / 11887 | 20 / 20 |
| gpt-6.1-terra | 3786 / 6967 | 3786 / 6967 | 6444 / 8405 | 20 / 20 |
| gemini-3.8-flash | 8481 / 10952 | 8481 / 10952 | 9518 / 11313 | 20 / 20 |
How to read it: "first token" is the first streamed chunk of any kind. grok-4.7 and deepseek-v4.1-flash stream reasoning text before the answer, so their first answer text arrives later than their first token. 2 of the 100 requests ended normally but returned no answer text: one claude-sonnet-5-5 run, and one deepseek-v4.1-flash run that spent its 800 tokens on reasoning. We did not investigate the cause. With 20 samples per model, p95 is a rough guide only. This is one location and one 11-minute window, so your numbers will differ.
Call all of them with one loop
from openai import OpenAI
client = OpenAI(
api_key="YOUR_TOKENLAB_KEY",
base_url="https://api.tokenlab.sh/v1",
)
for model in ["gpt-6.1-terra", "claude-sonnet-5-5", "grok-4.7", "gemini-3.8-flash"]:
resp = client.chat.completions.create(
model=model,
messages=[{"role": "user", "content": "One sentence on HTTP/2."}],
max_tokens=800,
)
print(model, resp.choices[0].message.content)Only the model string changes. Use the IDs as listed in the TokenLab model catalog.
How to test it
Add $5 to $10 of balance, point one non-critical job at the endpoint, and compare cost per finished task and error rate with your current setup. If the numbers do not work for you, stop; nothing else changes. Create a key or see live pricing. Prices can change; the live pages win.
Sources: TokenLab pricing API (registry revision 2026-11-04); openrouter.ai/api/v1/models/<model>/endpoints, read 2026-11-04; our own timing run described above. Written by TokenLab's AI team.
