Price-Rung Persistence
Instrument retired 2026-08-19. This concept is built on the Token Price Index, which was a swept series and is now a closed archive at
/datasets/token-prices. Everything below stands as dated analysis — the observations are primary, verified against providers' own pricing pages on the dates given, and remain citable. What no longer holds is the forward claim: a frontier rung cut will not "show up in the TPI on the day it happens", because nothing is sweeping it. Reading a future rung move now means going to the provider's pricing page directly.The page is deliberately left otherwise intact rather than rewritten. Its 2026-08-10 partial falsification is one of the better entries in this KB, and an instrument being retired does not make the analysis it produced wrong.
Ruled 2026-08-23: this concept is kept as a dated archive. Not re-grounded on another instrument, not closed. It is finished work, and its value is now historical — it records a stated call, then shows that call partly breaking, without quietly editing itself. Re-grounding it would mean hand-recording rung rows from each provider's pricing page at every release event, which is exactly the standing maintenance tax the retirement existed to shed. Nothing further is owed on this page. Do not open it as work again; a future rung observation is a new dated entry elsewhere, not a revival of this one.
Price-Rung Persistence
Frontier labs price by rung, not by cost. A capability tier has a nominal per-token price point — a rung on a ladder — and a new model generation steps onto the existing rung rather than lowering it. Capability improves at a fixed published price; the price does not fall for the buyer who stays at that tier. This is the sell-side complement to the price-decline distribution: the same behaviour that holds the rung steady is what forces the price of yesterday's capability to collapse.
⚠ PARTLY FALSIFIED — the strong rung broke on 2026-07-30 (recorded 2026-08-10)
The frontier rung held. The strong rung did not, and this page's headline example is the one that broke. OpenAI cut GPT-5.6 Terra by 20% — from $2.50/$15 to $2.00/$12 — and Luna by 80% — from $1.00/$6.00 to $0.20/$1.20. Both cuts are permanent, not promotional. Observed into the Token Price Index on 2026-08-03 (primary, OpenAI's own pricing page); the 30 July effective date is from secondary coverage.
2026-07-16 (this page's claim) 2026-08-03 (observed) Prior-gen incumbent, 2026-08-03 Sol (frontier) $5.00 / $30.00 — exact hold $5.00 / $30.00 — still holding GPT-5.5, $5.00 / $30.00 Terra (strong) $2.50 / $15.00 — "exact hold" $2.00 / $12.00 — rung broken GPT-5.4, still $2.50 / $15.00 Luna (mid) $1.00 / $6.00 — new higher rung $0.20 / $1.20 — rung collapsed GPT-5.4-nano, $0.20 / $1.25 Why this is a sharper falsification than the one the page anticipated. Open Question 3 below asked when a provider would "break the pattern and cut a frontier rung outright" — i.e. it expected the break to arrive as a next generation stepping onto a lower rung. What actually happened is worse for the thesis: the same model cut its own price fourteen days after launch, and it now sits below the prior-generation incumbent it was said to have exactly matched. GPT-5.4 is unchanged at $2.50/$15 while its successor undercuts it by 20%. The rung is not the fixed object; at the strong tier the rung moved and the incumbent was left stranded above it. Luna's collapse also destroys the "new, higher mid rung" reading — the mid tier did not get a higher rung, it got an 80% cut inside three weeks.
What survives. The frontier rung held, so the strategic claim is narrowed to the frontier tier, not destroyed. The mechanism the page proposes for stickiness (scarce serving capacity rationed by price) is untouched by a competitive cut one tier down — and DeepSeek's 2026-08-06 notice of a "significant" price rise, driven by demand it could not serve, is consistent with it. Two providers moved in opposite directions in the same week.
What this does to the solvency reading. The page argues that "sticky premium pricing" is one of two joint conditions in Matsuoka's corridor, and that a rung cut is the leading indicator that half the corridor has broken. That indicator has now fired at the strong tier. The page also claimed the TPI would show it "on the day it happens" — it did not. The cut landed 30 July, was observed 3 August, and sat unread until 10 August because the CSV row was labelled
NEW to indexon a slug already in the index, so the sweep's delta report had nothing to raise. The instrument worked; the alerting on top of it did not.Nothing below this box has been rewritten. The original claim is the record, and a page that quietly edited itself would be worth less than one that shows the call being broken.
The evidence (own dataset, primary against OpenAI's pricing page, 2026-07-16). OpenAI released GPT-5.6 in three variants on 2026-07-16. Recorded into the MenFem Token Price Index against the prior-generation rows (verified 2026-07-06), two of three variants land on the exact existing rung:
| GPT-5.6 variant | Tier | Input / Output (per Mtok) | Prior-gen rung | Result |
|---|---|---|---|---|
| Sol | frontier | $5.00 / $30.00 | GPT-5.5 — $5.00 / $30.00 | exact hold |
| Terra | strong | $2.50 / $15.00 | GPT-5.4 — $2.50 / $15.00 | exact hold |
| Luna | mid | $1.00 / $6.00 | GPT-5.4-mini — $0.75 / $4.50 | new point (above prior mid) |
A capability generation (GPT-5.5 → GPT-5.6 Sol; GPT-5.4 → GPT-5.6 Terra) shipped at the identical published per-token price, to the cent. The price point is the fixed object; the model that occupies it got better.
Luna is the honest nuance — the rule is not a law. The new mid variant at $1/$6 is a new rung, priced above OpenAI's prior mid point (GPT-5.4-mini at $0.75/$4.50), not a held one. So "the rung never moves" describes the frontier and strong tiers cleanly this cycle, but the mid tier got a new, higher rung. The pattern is two-for-three, not three-for-three — and one provider, one release event, is a data point, not a proven law.
Why it matters for the margin question. If the rung holds while capability rises, the provider captures the entire capability gain as margin (or as relief from its own falling unit cost) rather than passing it to the buyer as a lower price. That is the opposite of what commodity infrastructure does. It is also exactly what makes the price-decline distribution so steep for fixed capability: a buyer who only needs GPT-5.5-level performance sees its price fall not because the frontier rung dropped, but because that capability is now available a rung down (Terra) or from open-weight models. The rung holding at the top is the engine of the collapse below it.
The rung is now a named solvency variable, not just a pricing curiosity (added 2026-07-22). Matsuoka's preprint (arXiv 2607.07207, 2026-07-08) models the solvency of the announced AI buildout as a narrow corridor requiring two things jointly: ~2x annual token-demand growth for four years, AND sticky premium pricing. "Sticky premium pricing" is precisely what this page measures. When OpenAI held the frontier rung at $5/$30 and the strong rung at $2.50/$15 across a capability generation on 2026-07-16, that was one of the two corridor conditions being satisfied in public, on a dated, primary-observable basis.
This upgrades the Token Price Index from a pricing dataset to an instrument on a solvency variable. The practical consequence: a frontier rung cut, whenever it comes, is the leading indicator that half the solvency corridor has broken — and it will show up in the TPI on the day it happens, months before it shows up in anyone's financials. Two caveats keep this honest: Matsuoka's corridor is modelled, not measured (single-author preprint, inputs undisclosed); and a held list price is not a held realised price — enterprise discounting, committed-use contracts and batch/cached tiers are invisible to this instrument. See the cost-measurement problem for the general form of that blind spot.
A cost-side reason the rung can stay sticky (added 2026-07-23). Epoch's serving-capacity model (2026-05-25, calibrated to 111 InferenceX runs) finds global serving capacity growing ~3.4x/yr against demand growing ~10x/yr — a supply-demand gap. Scarce serving capacity rationed by price is a mechanistic reason a provider can hold a nominal rung even as the cost of fixed capability falls fast beneath it: when you cannot serve all comers, you do not have to cut price to win the marginal buyer. This is a cost/capacity-side complement to the strategic reading above, not a substitute for it.
Two-of-three at one release is not a curve — but a multi-model curve now exists alongside it (added 2026-07-23). This page rests on one provider, one event (GPT-5.6). Tiered Super-Moore's Law (Du, 2603.28576) supplies the complement: tier-level price half-lives across many models — economy ~1.10yr, mid ~1.55yr, flagship near-zero-fit (reasoning premium ~31.5x). It measures the rate the price of a tier falls over time across the market; the rung observation measures whether a specific provider steps a new generation onto the old price point. They are different cuts (market-wide curve vs provider generation-over-generation rung), and Du's data does not give the generation-over-generation rung rows this page still lacks for Anthropic / Google / DeepSeek. The gap on this page is narrowed in context, not closed.
Key Claims
- GPT-5.6 Sol (frontier) = $5/$30 — the exact GPT-5.5 frontier rung. Evidence: strong (own dataset, primary vs OpenAI's pricing page 2026-07-16) (TPI)
GPT-5.6 Terra (strong) = $2.50/$15 — the exact GPT-5.4 strong rung.FALSIFIED 2026-07-30, recorded 2026-08-10. Terra cut to $2.00/$12 while GPT-5.4 remains at $2.50/$15 — the successor now undercuts the incumbent by 20%. Evidence: strong (own dataset, primary, observed 2026-08-03)GPT-5.6 Luna (mid) = $1/$6 — a new mid point, above the prior GPT-5.4-mini rung ($0.75/$4.50).FALSIFIED 2026-07-30, recorded 2026-08-10. Luna cut 80% to $0.20/$1.20, now below GPT-5.4-nano on output at the same input rung. The "new higher mid rung" lasted three weeks. Evidence: strong (own dataset, primary, observed 2026-08-03)Two of three tiers held the prior-generation rung to the cent.Now one of three. Only Sol (frontier) still holds at $5/$30 as of 2026-08-03. Evidence: strong (own dataset)- "Sticky premium pricing" is one of two joint conditions for buildout solvency (the other: ~2x annual token-demand growth for four years). Rung-holding is therefore a directly observable solvency signal — and the signal has now fired at the strong tier, though not at the frontier. Evidence: weak-moderate (modelled corridor, single-author preprint) (Matsuoka)
- A rung can break from within a generation, not only across one. Terra cut its own price 14 days after launch. The page's own Open Question 3 anticipated the break arriving as a next generation stepping lower; that framing was too narrow. Evidence: strong (own dataset, primary)
- The pattern is tier-specific, not provider-specific — the first cross-provider evidence points the same way. Anthropic's generation shows OpenAI's exact shape: the frontier rung held to the cent (Opus 5 at $5/$25, identical to Opus 4.8) while the strong rung broke (Sonnet 5 at $2/$10 against Sonnet 4.6's $3/$15). Two providers is not an industry, but "frontier holds, strong breaks" now survives a provider change, which "OpenAI holds its rungs" does not. Evidence: moderate-strong (own dataset for the Anthropic prices, primary vendor pages for both; n=2 providers, Google unobserved)
- List-price stickiness is not realised-price stickiness — discounts, committed-use contracts and batch/cached tiers are invisible to the Token Price Index. Evidence: strong (a known limit of the instrument, not a source claim)
Benchmarks & Data
At release, 2026-07-16 — the original observation:
| Tier | GPT-5.6 price | Prior-gen price | Held? |
|---|---|---|---|
| frontier | $5.00 / $30.00 (Sol) | $5.00 / $30.00 (GPT-5.5) | yes, exact |
| strong | $2.50 / $15.00 (Terra) | $2.50 / $15.00 (GPT-5.4) | yes, exact |
| mid | $1.00 / $6.00 (Luna) | $0.75 / $4.50 (GPT-5.4-mini) | no — new, higher point |
Eighteen days later, 2026-08-03 — the same rows re-observed:
| Tier | GPT-5.6 price | Change since release | Prior-gen incumbent | Rung intact? |
|---|---|---|---|---|
| frontier | $5.00 / $30.00 (Sol) | unchanged | GPT-5.5 — $5.00 / $30.00 | yes |
| strong | $2.00 / $12.00 (Terra) | −20% / −20% | GPT-5.4 — still $2.50 / $15.00 | no — cut below the incumbent |
| mid | $0.20 / $1.20 (Luna) | −80% / −80% | GPT-5.4-nano — $0.20 / $1.25 | no — collapsed |
Source: MenFem Token Price Index (research/inference/token-prices.csv), rows observed 2026-07-06, 2026-07-16 and 2026-08-03, each verified against OpenAI's own pricing page. The 30 July effective date for the cuts is secondary; the 2026-08-03 observation is primary.
The first Anthropic observation, 2026-08-13 — Open Question 1's first real answer:
| Tier | Anthropic new-gen price | Prior-gen incumbent | Rung intact? |
|---|---|---|---|
| frontier | $5.00 / $25.00 (Opus 5) | Opus 4.8 — $5.00 / $25.00 | yes, exact |
| frontier, new rung above | $10.00 / $50.00 (Fable 5) | none — a new top rung | n/a — the Luna move at the top of the ladder rather than the bottom |
| strong | $2.00 / $10.00 (Sonnet 5) | Sonnet 4.6 — $3.00 / $15.00 | no — launched a third below it, and the cut was made permanent 2026-08-10 |
Anthropic broke the strong rung by a different mechanism than OpenAI, and it is the sharper case. Terra simply cut. Anthropic launched Sonnet 5 below the incumbent rung on 2026-06-30 but labelled the cut introductory, with a return to the $3/$15 rung scheduled for 1 September — the rung restoration was explicit, dated and published. On 2026-08-10 it cancelled that restoration and made $2/$10 the standard price. A provider that puts a return to the rung on the calendar and then abandons it is stronger evidence against rung-holding than one that never announced an intention to hold: here the intent existed, was public, and was withdrawn.
One qualification, and it cuts against the size of the break rather than its direction: Sonnet 5 bills ~30% more tokens for the same text than Sonnet 4.6 (new tokenizer; Sonnet 4.6 is on the old one). Per unit of text the strong rung fell about 13%, not 33% — see Inference & Token Economics. The rung broke, by less than the price sheet says.
Source: Opus 5, Opus 4.8 and Sonnet 5 prices from the Token Price Index (rows observed 2026-08-03, verified against Anthropic's own pricing page). Sonnet 4.6's $3/$15 and the cancellation were read directly from the Anthropic pricing page and its 2026-08-10 release-notes entry on 2026-08-13 — Sonnet 4.6 is not in the Index, which is a gap: the instrument cannot check a rung whose incumbent it never recorded.
Open Questions
Is rung-holding an OpenAI convention or an industry-wide one?Partly answered 2026-08-13 — Anthropic, and it matches: frontier holds exactly, strong breaks (table above). The claim that survives a provider change is the tier-specific one, not the provider-specific one. Still open at n=2: Google/Gemini generations remain unobserved, and the Index does not yet carry prior-generation incumbents for either non-OpenAI provider.- Does the mid tier keep getting new, higher rungs (Luna) while frontier/strong hold — i.e. is the ladder growing rungs at the bottom rather than lowering them?
- When (if ever) does a provider break the pattern and cut a frontier rung outright — and what does that signal about competitive pressure at the top? (Under Matsuoka's corridor, that event is also a solvency signal, not just a competitive one.)
- How wide is the gap between list rungs and realised prices? Enterprise discounting could be quietly cutting the effective rung while the published one holds — which would make the instrument read the wrong direction.
- Does rung-holding survive contact with open-weight parity? If a frontier-capable open-weight model (Matsuoka cites GLM-5.2) sets a free floor one rung down, holding the paid rung above it is a different, harder act than holding it against a paid competitor.
Related Concepts
- The Price-Decline Distribution — the buy-side mirror: holding the rung while capability rises is what makes the price for fixed capability fall so fast.
- Token Pricing & the Inference-Margin Question — a held rung is evidence against the commodity-infrastructure reading, at least at the moment of a release.
- The Depreciation Conveyor & Vintage Economics — "sticky premium pricing" is one of that model's two solvency conditions; this page is the instrument on it.
- The Cost-Measurement Problem — the general form of this page's blind spot: published prices are observable, realised prices and costs are not.
Backlinks
Pages that reference this concept:
Changelog
- 2026-08-13 — FIRST CROSS-PROVIDER EVIDENCE. Open Question 1 ("is rung-holding an OpenAI convention or an industry-wide one?") gets its first real answer, and it is the same shape as OpenAI: Anthropic held the frontier rung exactly (Opus 5 = Opus 4.8 at $5/$25) and broke the strong rung (Sonnet 5 at $2/$10 against Sonnet 4.6's $3/$15). Occasioned by a correction elsewhere: Anthropic had scheduled Sonnet 5's introductory rate to step back up to the $3/$15 rung on 1 September and cancelled that step-up on 2026-08-10, which is the load-bearing fact — a published, dated intent to restore a rung, then withdrawn. Added a Key Claim (the surviving generalization is tier-specific, not provider-specific), the Anthropic table, and the tokenizer qualification (~13% not 33% per unit of text). Logged a gap the observation exposed: the Index carries no Sonnet 4.6 row, so the incumbent side of this comparison had to be read off the vendor page rather than the instrument.
- 2026-08-10 — PARTLY FALSIFIED. OpenAI cut Terra 20% ($2.50/$15 → $2.00/$12) and Luna 80% ($1/$6 → $0.20/$1.20) on 2026-07-30, permanently. Observed into the TPI 2026-08-03; not read here until today. The frontier rung (Sol, $5/$30) still holds, so the claim narrows to the frontier tier rather than dying. Terra now sits below GPT-5.4, which is unchanged at $2.50/$15 — a successor undercutting its own incumbent, which is a sharper break than the next-generation-steps-lower case Open Question 3 anticipated. Original claims struck through, not deleted. Also recorded: the 7-day delay was an alerting failure, not an instrument failure — the CSV row carrying the cut was labelled
NEW to indexon a slug already in the index, so the sweep's delta report had nothing to raise. - 2026-07-23 — Light-touch additions from the discovery sweep: Epoch's ~3.4x-supply / ~10x-demand serving-capacity gap as a cost/capacity-side reason a rung can stay sticky; and Du's multi-model tier half-lives (economy 1.10yr / mid 1.55yr) as the market-wide-curve complement to this page's single-provider generation-over-generation rung observation — narrowing but not closing the "one provider, one event" gap. No change to the GPT-5.6 rung data.
- 2026-07-22 — Compiled Matsuoka (2607.07207) against this page: "sticky premium pricing" identified as one of two joint solvency conditions in his corridor model, which reframes the Token Price Index as an instrument on a solvency variable and a frontier rung cut as a leading indicator. Added the list-price-vs-realised-price limit of the instrument.
- 2026-07-16 — Created from the GPT-5.6 release, observed into the Token Price Index the same day (own dataset, primary vs OpenAI's pricing page). Sol/Terra held the prior-gen rung exactly; Luna is a new mid point (the disclosed exception). First sell-side concept in this topic; paired with price-decline-distribution.