Touchstone Assay
Dated SIU — SIU-2026a
$0.0111
25 Sep 2026 provisional
Correction notice — published after this print, disclosing a fact that does not change dated_siu:

4 corrections: 1 precision, 2 pricing, 1 other

  • 25 Sep 2026 — dated_siu previously rounded to a fixed 4 decimal places rather than 4 significant figures — found live and fixed the same day, 2026-09-25 (docs/methodology.md's Rounding section). At this print's price level that was too coarse to show real day-to-day movement. This print's signed body is never edited — the published figure (0.0111) stands. Recomputed from this print's own published basket_costs and weights (Σ weight × basket_cost, rounded under the corrected 4-significant-figure rule, reproducible by anyone from this print's own published figures alone), the real value is 0.01111.
  • 25 Sep 2026 — This print was published under the belief that cached-input pricing (rules-2026-09-13, commit 29d4be2) was already live — docs/methodology.md stated this as fact. It was not: buildModelInputs never forwarded a price snapshot's price_cached_in_usd_per_1m to the cost formula, for any model, from any source, until this was found and fixed live 2026-09-25 (commit 327d7b9, docs/methodology.md's rules-2026-09-25b). For this print specifically, real nonzero cached_input usage on deepseek-v3.2, gemini-3.1-pro-preview, grok-4.6, mistral-small-3.2-24b-instruct — already present in this print's own run records — was still priced at zero in this print's own signed cost_of_production_usd and basket_costs, contrary to what the methodology this print's date falls under claimed. This print's signed body is never edited. The exact, per-model corrected dollar figure is not computed here — that quantified backfill is separate, deferred follow-up work; this note discloses the structural fact, not yet the number.
  • 26 Sep 2026 — Quantified follow-up to this print's own cached-input-pricing correction note above (the "separate, deferred follow-up work" that note left open). Corrected figures, computed from this print's own real run records and price snapshot with the fix applied (commit 327d7b9), not from new data: this print's dated_siu was originally reported as $0.0111; the corrected figure is $0.0112. gemini-3.1-pro-preview's own basket cost was originally reported as $0.039236; the corrected figure is $0.040218. grok-4.6's own basket cost was originally reported as $0.017048; the corrected figure is $0.017289. deepseek-v3.2 and mistral-small-3.2-24b-instruct also had real, nonzero cached_input usage this day (per the original note), but it did not move either model's own basket cost at the published 6-decimal-place precision. This print itself is not recomputed or replaced — the original stands, per the revision policy's unconditional never-edit guarantee. These are also now the exact figures packages/print's own verify command (recomputing under this print's own historical rounding rule) reports.
  • 27 Sep 2026 — claude-haiku-4-5 produced zero run records this print (excluded_reason: no run records for T1/T2/T3): all 15 real calls failed with a genuine HTTP 400, "Your credit balance is too low to access the Anthropic API" — a real rejection, not a transient provider failure. Same cause and mechanism as the equivalent note on 2026-09-17: claude-sonnet-5 runs first in registry order, its own 15 calls succeeded (real cost /bin/bash.023499, this print's own basket_costs) and consumed the last of a low prepaid balance; every one of haiku's own calls then hit a genuinely empty balance. Not a provider outage, not a harness bug — confirmed against this project's own real Anthropic invoice history, cross-referenced with the real run log. See docs/methodology.md's billing-exhaustion disclosure.
▲ +2.8% since 24 Sep 2026 — More expensive than the previous print.
Equal weighting, disclosed: basket costs span 59.8× across the 9 qualifying constituents. 3 of 9 constituents (33%) sit at or above the mean basket cost and together account for 80% of the summed basket cost Dated SIU is computed from — equal weighting still gives each of them the same 1÷9 nominal share. Dated SIU therefore tracks the more expensive end of the basket more closely than a demand-weighted index would. See the methodology's Weighting section, and Frontier SIU / Commodity SIU, the tier prints published alongside this one for exactly this reason.

Exchange-rate table

ModelUSD / SIUSpread to indexSIU per $1
deepseek-v3.2 $0.0018 -84% 547.9
llama-3.3-70b-deepinfra $0.0007 -94% 1410.3
mistral-small-3.2-24b-instruct $0.0007 -94% 1525.2
qwen-2.5-72b-instruct $0.0025 -77% 396.8
claude-sonnet-5 $0.0235 +112% 42.5
gpt-5.1 $0.0091 -18% 109.4
gemini-3.1-pro-preview $0.0392 +253% 25.4
grok-4.6 $0.0170 +53% 58.6
claude-haiku-4-5 — — —
undefined class: T1 (no run records for this class); T2 (no run records for this class); T3 (no run records for this class)
gpt-5.4-mini $0.0053 -52% 187.0

Floor & market spread

Not yet published — the floor needs a measured hardware-rental rate for the reference configuration, which this print does not yet include.

Sensitivity

The print recomputed under alternative cache and batch-discount policies, so the effect of those policy choices is stated rather than hidden inside one number.

Policy variantDated SIUDeltaApplies to
cache-enabled: 40% of T2 input at 10% of input price $0.0080 -27.8% deepseek-v3.2, llama-3.3-70b-deepinfra, mistral-small-3.2-24b-instruct, qwen-2.5-72b-instruct, claude-sonnet-5, gpt-5.1, gemini-3.1-pro-preview, grok-4.6, claude-haiku-4-5, gpt-5.4-mini
batch-discount: 50% off input and output $0.0056 -50.0% deepseek-v3.2, llama-3.3-70b-deepinfra, mistral-small-3.2-24b-instruct, qwen-2.5-72b-instruct, claude-sonnet-5, gpt-5.1, gemini-3.1-pro-preview, grok-4.6, claude-haiku-4-5, gpt-5.4-mini

Methodology, signature & anchor

Methodology
v0-draft
Cost of production
$1.5941
Signature
0x778db23e…a58e6b9b
Public key
0x030c598e…e25cb299
Anchor tx
0x16f86419…fcb2a118 (base-sepolia)
Raw runs
https://github.com/petrkrulis2022/dated-siu-inference/tree/main/data/runs/2026-09-25

Dated SIU is an assessment of qualifying inference work under this basket version, model/provider configuration and quality gate — not a prediction of production agent costs, which vary with workflow design, tool reliability, retry policy and context handling.

Verify this yourself

Don't take this page's word for it — every value below is checkable independently.

Body hash
0x20ab94ba…88689fc4
Recovered signer
0x284ff2f8…d48f845a — matches the TouchstoneAttestation publisher below
TouchstoneAttestation
0xF6070179…93297190 (base-sepolia)
  1. Recompute this print's body hash from the published JSON (JCS-canonicalised, signature fields excluded) and confirm it matches the body hash shown above.
  2. Open the contract link above on Blockscout, go to "Read Contract", call publisher(), and confirm it matches the recovered signer shown above.
  3. On the same "Read Contract" tab, call postedAt(bytes32) with the body hash shown above and confirm it returns a non-zero timestamp.

All prints