Published 07:13 AEST · Before US Open · Wednesday, 9 September 2026Pipeline 04:30 AESTVol. I  No. 121
The Grid
Independent Price Reporting for the Compute Market
thegridco.ai
EST. 2026 · SYDNEY
The normalisation ladder · rung 3

US$ per million tokens

What a million tokens costs at a provider's published list price. The denomination hosted inference is sold in — and the rung where model choice starts to hide inside the number.

Reference — stated, not assessed  ·  notice 2026-08-20 §2

· The honesty frame

a reference table · not an assessment
The honesty frame

What this table is. The per-token list prices providers publish, recorded daily from two documented public API endpoints (source ratified by founder ruling 2026-08-20; one request per endpoint per daily cycle) and rendered here as stored. A record of what is published — with source, method, observation count and date.

What it is not. Not an assessment and not an indicator: no adjustment, no judgment, no panel, no median — and no claim that any listed price is the price a volume buyer pays.

Scope, stated. This table renders the two-catalogue collection only. The same store also holds first-party price-card rows from a separate collection; those enter the Reference Workload Index under its own specification — see /standards/instruments/rwi— and are out of scope for this rung’s table.

The denomination’s own limitation. Models differ in how many tokens they spend on the same work — on the order of 2× — so a cheaper $/MTok is not a cheaper task. That limitation is why rung 4 is declined, and why the measured-throughput column below is a stated gap rather than a borrowed number.

· Published list prices, as recorded

63 rows · collection date 2026-09-09
Method: tracked-model allowlist · list prices as published, USD per million tokens · zero- and free-tier listings excluded · rows land as collected, rendered without adjustment.
DeepInfra — documented public API (models/list) · 14 rows
Model (as listed)Input $/MTokOutput $/MTokMeasured throughput
deepseek-ai/DeepSeek-V4-Flash$0.090$0.180[gap: licence pending]
deepseek-ai/DeepSeek-V4-Flash-0731$0.060$0.180[gap: licence pending]
deepseek-ai/DeepSeek-V4-Flash-Vision-Exp$0.440$1.320[gap: licence pending]
deepseek-ai/DeepSeek-V4-Pro$1.300$2.600[gap: licence pending]
deepseek-ai/DeepSeek-V4-Pro-0813$1.300$2.600[gap: licence pending]
meta-llama/Llama-3.3-70B-Instruct-Turbo$0.100$0.320[gap: licence pending]
moonshotai/Kimi-K2.6$0.750$3.500[gap: licence pending]
moonshotai/Kimi-K2.7-Code$0.680$3.400[gap: licence pending]
moonshotai/Kimi-K3$2.850$14.25[gap: licence pending]
Qwen/Qwen3.5-122B-A10B$0.290$2.400[gap: licence pending]
Qwen/Qwen3.5-27B$0.260$2.600[gap: licence pending]
Qwen/Qwen3.5-35B-A3B$0.140$1.000[gap: licence pending]
Qwen/Qwen3.5-397B-A17B$0.450$3.000[gap: licence pending]
Qwen/Qwen3.5-9B$0.100$0.150[gap: licence pending]
OpenRouter — documented public API (api/v1/models) · 49 rows
Model (as listed)Input $/MTokOutput $/MTokMeasured throughput
anthropic/claude-fable-5$10.00$50.00[gap: licence pending]
anthropic/claude-fable-5:batch$5.000$25.00[gap: licence pending]
anthropic/claude-fable-5.1$10.00$50.00[gap: licence pending]
anthropic/claude-fable-5.1:batch$5.000$25.00[gap: licence pending]
anthropic/claude-opus-4.8$5.000$25.00[gap: licence pending]
anthropic/claude-opus-4.8:batch$2.500$12.50[gap: licence pending]
anthropic/claude-sonnet-5$2.000$10.00[gap: licence pending]
anthropic/claude-sonnet-5:batch$1.000$5.000[gap: licence pending]
deepseek/deepseek-v4-flash$0.070$0.140[gap: licence pending]
deepseek/deepseek-v4-flash-0731$0.065$0.180[gap: licence pending]
deepseek/deepseek-v4-flash-0731:batch$0.110$0.330[gap: licence pending]
deepseek/deepseek-v4-flash-vision-exp$0.220$0.660[gap: licence pending]
deepseek/deepseek-v4-flash-vision-exp:batch$0.110$0.330[gap: licence pending]
deepseek/deepseek-v4-pro$0.870$1.740[gap: licence pending]
deepseek/deepseek-v4-pro-0813$0.579$1.738[gap: licence pending]
deepseek/deepseek-v4-pro-0813:batch$0.660$1.980[gap: licence pending]
google/gemini-3.1-pro-preview$2.000$12.00[gap: licence pending]
google/gemini-3.1-pro-preview-customtools$2.000$12.00[gap: licence pending]
google/gemini-3.1-pro-preview:batch$1.000$6.000[gap: licence pending]
meta-llama/llama-3.3-70b-instruct$0.100$0.320[gap: licence pending]
moonshotai/kimi-k2$0.570$2.300[gap: licence pending]
moonshotai/kimi-k2-0905$0.600$2.500[gap: licence pending]
moonshotai/kimi-k2-thinking$0.600$2.500[gap: licence pending]
moonshotai/kimi-k2.5$0.450$2.250[gap: licence pending]
moonshotai/kimi-k2.6$0.950$4.000[gap: licence pending]
moonshotai/kimi-k2.7-code$0.710$3.500[gap: licence pending]
moonshotai/kimi-k3$3.000$15.00[gap: licence pending]
moonshotai/kimi-k3:batch$3.000$15.00[gap: licence pending]
openai/gpt-5.6-luna$0.200$1.200[gap: licence pending]
openai/gpt-5.6-luna-pro$0.200$1.200[gap: licence pending]
openai/gpt-5.6-luna-pro:batch$0.100$0.600[gap: licence pending]
openai/gpt-5.6-luna:batch$0.100$0.600[gap: licence pending]
openai/gpt-5.6-sol$2.000$10.00[gap: licence pending]
openai/gpt-5.6-sol-pro$2.000$10.00[gap: licence pending]
openai/gpt-5.6-sol-pro:batch$1.000$5.000[gap: licence pending]
openai/gpt-5.6-sol:batch$1.000$5.000[gap: licence pending]
openai/gpt-5.6-terra$2.000$12.00[gap: licence pending]
openai/gpt-5.6-terra-pro$2.000$12.00[gap: licence pending]
openai/gpt-5.6-terra-pro:batch$1.000$6.000[gap: licence pending]
openai/gpt-5.6-terra:batch$1.000$6.000[gap: licence pending]
qwen/qwen3.5-122b-a10b$0.290$2.400[gap: licence pending]
qwen/qwen3.5-27b$0.195$1.560[gap: licence pending]
qwen/qwen3.5-35b-a3b$0.313$1.250[gap: licence pending]
qwen/qwen3.5-397b-a17b$0.550$3.500[gap: licence pending]
qwen/qwen3.5-9b$0.100$0.150[gap: licence pending]
qwen/qwen3.5-9b:batch$0.170$0.250[gap: licence pending]
qwen/qwen3.5-flash-02-23$0.065$0.260[gap: licence pending]
qwen/qwen3.5-plus-02-15$0.260$1.560[gap: licence pending]
qwen/qwen3.5-plus-20260420$0.300$1.800[gap: licence pending]
Measured throughput — the gap is displayed, not hiddenTokens-per-second measurement across providers exists behind a commercial licence The Grid has not purchased (Artificial Analysis; its terms bar scraping and its free API tier is licensed for internal use only). Every cell that feed would fill renders [gap: licence pending] until a paid licence exists — methodology change notice 2026-08-20 §4.
Rung 3 of the normalisation ladder (notice 2026-08-20-normalisation-ladder, 2026-08-20). Reference class: published list prices as recorded, never assessed. Model names render as the venues list them; a listing here is a record of publication, not an endorsement or a comparison.