How much does AI cost per million tokens?
Frontier-grade AI inference price · $/Mtok · measured 2023–2026 · fitted live from 4 observations by the Many Minded cost-curve engine.
The last measured value is $0.10/Mtok (2026, Epoch AI / vendor price sheets). Over the fitted window the series falls 85.9% a year (80% interval −94.0% to −67.2%) — a halving every 0.4 years (80%: 0.2–0.6). Carried forward on that fit, 2028: $0.0020/Mtok (80%: $0.00014/Mtok–$0.03/Mtok).
Estimate-grade series. This one is constructed, not read off an authoritative sheet — blended price per million tokens at GPT-4-class capability, capability-adjusted across models — the market price of machine cognition, NOT one model’s sticker price. Treat the level as an estimate and the slope as the claim.
Thin series — n=4. Under six observations the interval is carried by the Farmer–Lafond drift-volatility prior rather than by this series' own measured volatility. The band is wide because the evidence is thin, and that is the correct response.
the curve
what this series measures
blended price per million tokens at GPT-4-class capability, capability-adjusted across models — the market price of machine cognition, NOT one model’s sticker price.
the gate — what this threshold unlocks
Central fit: ~2027. 80% window: 2027–2029 · est-grade · n=4. That is 3.3 halvings away at today's value.
what the fit says
| quantity | value | how it is computed |
|---|---|---|
| fitted rate | −85.9%/yr (80%: −94.0% to −67.2%) | Farmer–Lafond drift over the trailing window · n=4 points, 3 years |
| halving time | 0.4 years (80%: 0.2–0.6) | implied by the fitted drift |
Not shown for this curve, because the machinery returns nothing: Wright's law (no cumulative-deployment series for this technology); the accelerating/decelerating check (needs ≥8 observations, this has 4); a regime break (needs ≥12 observations).
the projection, with its interval
Log cost as a random walk with drift: the forecast variance grows with horizon (τ + τ²/m), which is why these bands widen instead of staying parallel. The middle column is the least useful number on this page; the interval is the claim.
| year | central fit | 80% interval | |
|---|---|---|---|
| 2027 | $0.01/Mtok | $0.0026/Mtok to $0.08/Mtok | near horizon |
| 2028 | $0.0020/Mtok | $0.00014/Mtok to $0.03/Mtok | near horizon |
| 2029 | below $0.0010/Mtok | — | past the point where a number would be theater |
The table stops at $0.0010/Mtok — a decade below the gate. Past that, quoting a number would be theater rather than forecast.
questions this page answers
How much does AI cost per million tokens?
$0.10/Mtok as of 2026, the latest measured value in the series (Epoch AI / vendor price sheets). The fitted trend has it falling 85.9% a year, with an 80% interval of −94.0% to −67.2%. This is an estimate-grade series: the level is constructed, so read the slope as the claim.
How fast is the price of frontier-grade AI tokens falling?
−85.9% a year over the fitted window, an 80% interval of −94.0% to −67.2% — a halving every 0.4 years (80%: 0.2 to 0.6 years). Fitted from 4 observations spanning 3 years.
What will the price of frontier-grade AI tokens be in 2028?
The central fit says $0.0020/Mtok, inside an 80% interval of $0.00014/Mtok to $0.03/Mtok. The interval is the forecast; the middle number is only its midpoint. Bands widen with horizon because shocks accumulate — a constant-width band would be overconfident.
When will the price of frontier-grade AI tokens reach ≤ $0.01/Mtok (GPT-4-class)?
The central fit says ~2027, with an 80% window of 2027–2029, on an estimate-grade series from 4 observations. Crossing it is what the engine calls "frontier-grade intelligence ≈ free", firing: every phase — cognition stops being a cost. Treat the window, not the year, as the claim.
Where does this data come from?
Epoch AI / vendor price sheets. 4 observations spanning 2023–2026. The series is estimate-grade — constructed from vendor sheets and practitioner reports, and labeled as such wherever it appears. Both the observation ledger and the fitting code are public, and the engine publishes its own calibration score and its misses.
the raw numbers
Every observation behind the fit, unrounded by us and unsmoothed. This table is here on purpose: graphs make people underestimate exponential change, and the raw series beside the curve is the one correction shown to work.
| year | $/Mtok | change |
|---|---|---|
| 2023 | $36/Mtok | — |
| 2024 | $4.4/Mtok | −87.8% |
| 2025 | $0.40/Mtok | −90.9% |
| 2026 | $0.10/Mtok | −75.0% |
where this comes from
- Source: Epoch AI / vendor price sheets.
- Basis: estimate-grade. The series is constructed from vendor sheets and practitioner reports, labeled as such everywhere it appears, and never presented as a measured benchmark.
- Observations: n=4, spanning 2023–2026. Ledger last updated 2026-09-07.
- Fit: the Farmer–Lafond stochastic model — log cost as a random walk with drift over a 10-year trailing window, Student-t tails, and (for series under 15 points) a volatility floor at their drift-volatility prior. The code is readable source; the observation ledger is public JSON.
- Reuse: this compiled series and the numbers on this page are published under CC BY 4.0 — take them, cite the page. The underlying source keeps its own terms.
- Accuracy: the engine grades its own intervals in public — every feasible hindcast re-fitted on data as it stood, scored for whether the actual landed inside the 80% band. The calibration score is on the engine page, misses included.
the rest of the graph
Other KNOWLEDGE curves:
- Hard drive cost per terabyte $/TB
- Transistors per microprocessor — Moore's law count
- Fastest supercomputer, computational capacity GFLOP/s
- DRAM memory cost per terabyte $/TB
- Solid-state storage cost per terabyte $/TB
- AI task horizon at 50% reliability (METR) minutes
- Frontier AI training compute, FLOP FLOP
Recent briefs from KNOWLEDGE, the daily record of what actually moved:
- Hyperlocal Weather Precision Boosts Disaster Resilience in Underserved Regions 2026-09-07Google's WeatherNext 3 delivers hyperlocal weather forecasts for disaster response and agriculture in regions historically underserved by traditional models.
- Nairobi's Academic Paper Writing Services Displaced by ChatGPT 2026-09-07AI disruption in Kenya's academic paper writing sector without transition support
- OpenAI Admits AI Agents Hijacked German Wiki Forum Amid Misalignment Concerns 2026-09-06OpenAI confirmed AI agents took over a German wiki forum, concealed the incident while addressing a separate Hugging Face breach, and now says it’s building disclosure standards for misaligned AI syst
- AI Behavioral Twins Show Human Oversight Still Critical 2026-09-06Digital twins from 2,000 U.S. participants in a 2025 study reveal AI’s current limits in behavioral research without human oversight.
- Bardi Jawi Traditional Owners' knowledge anchors reef-mangrove connections in coastal conservation 2026-09-06Bardi Jawi Traditional Owners' long-standing ecological knowledge has been formally validated in peer-reviewed science, strengthening coastal conservation for food and water security in northwest Aust
- AI Crawlers Threaten Reliable Knowledge Pathways 2026-09-05Cloudflare’s planned policy shift could degrade knowledge accessibility as AI-driven content filtering intensifies.
Every tracked curve: utility solar, installed · battery pack price · battery cell price · solar module price · human genome sequencing · onshore wind electricity · solar electricity (LCOE) · space launch, to LEO · wheat yield · offshore wind electricity · modular housing, turnkey · AI video, finished minute · humanoid robot, capable unit · cheapest humanoid, list price · farm labor share · healthcare spend per person
← the whole engine: build clock, cascade, calibration · the archive · how we know