Skip to content
Slow Variables
Ask

Model

Price per unit of capability (cheapest METR-hour on OpenRouter)

For each price date, the lowest OpenRouter prompt price (USD per million tokens) divided by the model's latest METR 50% time horizon in hours, across models present in both datasets. Derived metric frontier_price_per_horizon_hour.

Reason through this

The reading

The series records price change-points from 10 September 2026; a direction needs several months of them.

Every readingeach dot links to its record
$0.30 · 2026-09-10 · measured
a readingthe rule needs 5 readings; 1 so far
measured

1 reading drawn from the derived metric frontier_price_per_horizon_hour; 2 observations in the evidence.

Derived (frontier_price_per_horizon_hour) from: METR time horizons (v1.1); OpenRouter model prices. METR, Measuring AI Ability to Complete Long Tasks (Horizon v1.1). OpenRouter list prices.

The numbers
as ofreadingvaluerecord
2026-09-10—$0.30measured

Why it matters

Price per unit of capability is the lab layer's capture proxy. Falling means the surplus from better models is passed to applications and users; flat or rising means labs keep it.

What cuts against it

One router's list prices, prompt tokens only; METR's horizon is software tasks under its suite, not general capability; the cheapest model per horizon-hour is a moving target as models are added.

How it is measured
  • derived metric frontier_price_per_horizon_hour (formula in the semantic layer)
  • source METR time horizons (v1.1) · default tier 1 · METR; analysis code MIT at https://github.com/METR/eval-analysis-public
  • source OpenRouter model prices · default tier 3 · Public API
  • proxy price · cadence daily

Four price change-points; a move under two cents per million tokens per horizon-hour is noise. Rising = labs keeping the surplus from capability gains.

Applied to metric:frontier_price_per_horizon_hour.

Derived rows (1)
as ofdimsvalueinputs
2026-09-10$0.30sources12
Latest readings
  • 2026-09-10 · $0.30as of 2026-09-10(2 obs)
Why this status, and its history

It has 1 reading and its direction rule compares 5, so it reads emerging rather than not yet measurable: the site uses not yet measurable only for what cannot be measured at all. The status turns into a direction once enough readings exist.

  1. 2026-09-27not yet measurable to emergingconf 40 → 40 · Claude (model)

    It has 1 reading and its direction rule compares 5, so it reads emerging rather than not yet measurable: the site uses not yet measurable only for what cannot be measured at all. The status turns into a direction once enough readings exist.

    Weighed against it: None needed: no reading has changed; this corrects the word used for too few readings.

    sources12
  2. 2026-09-10unmeasured to not yet measurableconf — → 40 · the evaluator (automatic)

    First price snapshot, 10 Sep 2026: the cheapest METR-measured capability on OpenRouter costs about $0.30 per million prompt tokens per hour of 50% time horizon (16 models present in both datasets). The series records price change-points, so a direction needs four more of them; not yet measurable until then. Initial seed.

    sources123456+26 more
Why this confidence

Rests on 2 observations from METR time horizons (v1.1), OpenRouter model prices; best evidence tier 1, grade A; current.

Confidence is independent of status: 90–95 multiple strong independent sources; 70–89 good evidence, some ambiguity; 50–69 mixed or hard to operationalise; 0–49 limited or vague evidence.

Predictions that lean on this