Frontier launch tracker
No model priced at $50 or more per million output tokens has kept a big share of OpenRouter spend. Claude Fable 5 and 5.1 peaked at 8.0% and 8.9% of spend in their first month; GPT-6 Astra, the one breakout, reached 22.8% and was down to 5.1% by week six. Claude Fable 5.2 is on day 3: 7.4% of spend over its first three days, against 6.8% for Fable 5.1 and 8.1% for GPT-6 Astra at the same point.
Claude Fable 5.2
listed Nov 9, 2026 · $8 / $40 per million tokensToo early for a verdict. The verdict needs four full weeks. Its best week so far is 7.4% of spend, against a bar of 15%.
Reported so far, outside OpenRouter
- List price, per million tokens in / out
- $8 / $4020% below Fable 5.1’s $10 / $50, the first cut to a Fable list price. Cached input is $0.20; fast mode is $16 / $80. Nov 9 · Anthropic · Anthropic pricing
- Artificial Analysis Intelligence Index, max effort
- 651st of 74 models on index v5.0.1: 5 points ahead of Gemini 4 Carbon, 6 ahead of Opus 5.5 and 15 ahead of Fable 5.1. Nov 9 · Artificial Analysis · AA by effort level · AA on Opus 5.5 · The Decoder
- Cost per Intelligence Index task, max effort
- $26.542.6 times Opus 5.5’s $10.04. At high effort Fable 5.2 scores 62, still ahead of Opus 5.5’s 59, for $9.08 a task. Nov 9 · AA by effort level · AA on Opus 5.5
- Output speed against Fable 5.1
- Up to 30% fasterAnthropic’s figure. A fast mode, up to 2.5 times faster again, costs twice the list price. Nov 9 · Anthropic
- Safeguard blocks against Fable 5.1
- About 55% fewerAnthropic’s early testing across its apps: about 60% fewer cybersecurity blocks and about 40% fewer biology blocks. Nov 9 · Anthropic
From launch day
Seven-day averages by day since each model was listed on OpenRouter; the first week averages the days so far. Dashed stretches include days the model was outside OpenRouter’s daily top 50 and had to be estimated. The 2026 flagship launches are in grey.
Fable 5’s clock stops while it was withdrawn (Jun 13 to Jun 30): its day 4 is July 1. Spend shares count each model at its list price; see the method.
Top of the benchmarks, bottom of the bill
Claude Fable 5 launched on June 9 at the top of Artificial Analysis’s Intelligence Index, about 5 points ahead of any other lab's best model, and took 16.1% of OpenRouter spend on its first full day. Three days after launch, Anthropic withdrew it to comply with US export controls. Back from July 1, it held 5.2% to 8.0% of spend a week and never more than 16% of Anthropic’s. Ramp’s card data on business spending tells the same story: about 11% of companies’ Anthropic spend in August, by then behind the cheaper Opus 5.
Fable 5.1 (September 1) kept the $10 / $50 price and cut the price of cached input by three-quarters. It peaked at 8.9% of spend in week 4, 30% of Anthropic’s. Then, on its day 21, came Opus 5.5, which Anthropic says matches Fable 5.1 on most tasks at $4 / $20. Opus 5.5 reached 33.4% of all OpenRouter spend in week 4; Fable 5.1 has since dropped out of OpenRouter’s daily top 50, which puts it under 4.1% of spend.
GPT-6 Astra, at the same $10 / $50, is the only premium launch to clear the 15% bar: 22.8% of spend in week 2 and 56% of OpenAI’s, enough to put OpenAI ahead of Anthropic on OpenRouter for three straight weeks from September 11, the first time in data back to January 2025. It faded as fast. On its day 18, OpenAI’s own GPT-6 Sol and Luna and Anthropic’s Opus 5.5 arrived; by week 6 Astra had 5.1% of spend, 22% of its peak.
Fable 5.2 (November 9) is the first Fable to cut the list price: $8 / $40, 20% below Fable 5.1. Artificial Analysis put it first on its Intelligence Index at 65: 5 points ahead of Gemini 4 Carbon, 6 ahead of Opus 5.5 and 15 ahead of Fable 5.1; $26.54 a task at max effort, against $10.04 for Opus 5.5. At high effort it scores 62 for $9.08 a task, less than Opus 5.5 costs at max. Anthropic’s own launch post tells users to start with Opus 5.5 for most coding and knowledge work, and to move up to Fable for bigger or harder jobs. On its first full day it took 9.1% of OpenRouter spend, against 7.0% for Fable 5.1, 8.4% for GPT-6 Astra and 16.1% for Fable 5.
The same thing has happened to every premium launch here. The premium model is the most capable available for a few weeks, then a cheaper model that is nearly as good arrives, often from the same lab, and takes the volume. Anthropic’s last premium tier, Opus 4 and 4.1 at $15 / $75 in 2025, peaked at 11.6% and 11.2% of spend; allowing for how much smaller the market was then, both reached $25 million faster than either Fable.
Days to $25 million
How long each launch took to reach $25 million of estimated spend through OpenRouter, the launch version of the time-to-100-million-users chart. Spending through OpenRouter is 2.5× what it was when Fable 5 launched, so the second view scales each day to today’s market: the model’s share of that day’s spending times an average day’s spending in the past week.
The game-changer test
Premium launches, priced at $40 or more per million output tokens. A game-changer clears the pull bar in its first four weeks and keeps 60% of it in week 6; one that clears the bar and loses it is a breakout that faded; one that never clears it is a frontier update. Upgrade and speed are shown for context. Weeks count from listing on OpenRouter.
| Fable 5.2 | Fable 5.1 | Fable 5 | GPT-6 Astra | Opus 4.1 ’25 | Opus 4 ’25 | |
|---|---|---|---|---|---|---|
| Listed | Nov 9 | Sep 1 | Jun 9 | Sep 4 | Aug 5 | May 22 |
| List price, per million tokens in / out | $8 / $40 | $10 / $50 | $10 / $50 | $10 / $50 | $15 / $75 | $15 / $75 |
| PullBest week in its first four, share of all OpenRouter spend (bar: 15%) | 7.4% · wk 1 … | 8.9% · wk 4 ✕ | 8.0% · wk 3 ✕ | 22.8% · wk 2 ✓ | 11.2% · wk 1 ✕ | 11.6% · wk 2 ✕ |
| UpgradeBest week in its first four, share of its own lab’s spend (bar: 40%) | 17% · wk 1 … | 30% · wk 4 ✕ | 16% · wk 4 ✕ | 56% · wk 4 ✓ | 16% · wk 2 ✕ | 18% · wk 3 ✕ |
| SpeedWeek 2, share of all OpenRouter spend (bar: 10%) | — … | 6.5% ✕ | 7.4% ✕ | 22.8% ✓ | 11.2% ✓ | 11.6% ✓ |
| Staying powerWeek 6, as a share of its best week (bar: 60%); only once it has cleared the pull bar | — … | n/a – | n/a – | ≈22% ✕ | n/a – | n/a – |
| Days to $25M of spend | not yet | ≈4127 at today’s size | ≈55≈24 at today’s size | 1614 at today’s size | never17 at today’s size | never18 at today’s size |
| Verdict | Too early | Frontier update | Frontier update | Breakout, faded | Frontier update | Frontier update |
≈ marks a value that includes estimated days. The bars are ours: 15% of spend is above every premium launch except GPT-6 Astra and below every Opus release since Opus 4.5; see the method.
Who carries each lab
Seven-day sums since May. Left: the share of each lab’s OpenRouter spend going to its premium models. Right: each lab’s share of all OpenRouter spend. Dashed stretches include estimated days.
Outside OpenRouter
OpenRouter is mostly developers and coding agents, and sees none of the traffic that goes to Anthropic or OpenAI directly. Business spending data from Ramp, the corporate card company, points the same way.
Business spend by model (Ramp)
- 1st. Claude Opus 4.828%
- 3rd. Claude Fable 58.0%
- 6th. Claude Opus 53.5%listed July 24
- Claude Fable 5~11%flat more than two months after launch; Opus 5 already ahead
- 1st. GPT-6 Astra19%
- 2nd. Claude Opus 517%
- 3rd. GPT-5.6 Sol11%
- 4th. Claude Opus 4.87.6%
- 5th. Claude Fable 5.16.6%
- 10th. Claude Fable 52.1%
Capability at launch (Artificial Analysis)
- Claude Fable 564.91st, about 5 points ahead of any other lab's best model; before September's revision, Jun 9. Artificial Analysis
- Claude Fable 5.153tied 1st, with GPT-6 Astra; $7.63 a task at max effort; revised index, Sep 9. Artificial Analysis
- GPT-6 Astra53tied 1st, with Fable 5.1; $3.26 a task at max effort; revised index, Sep 9. Artificial Analysis
- Claude Opus 5.5581st, 5 points ahead of GPT-6 Astra and Fable 5.1, at $4 / $20; revised index, Sep 22. OfficeChai
- Gemini 4 Argon53high effort; level with GPT-6 Astra; revised index, Oct 24. Artificial Analysis
- Claude Fable 5.2651st, 5 points ahead of Gemini 4 Carbon, 6 ahead of Opus 5.5 and 15 ahead of Fable 5.1; $26.54 a task at max effort, against $10.04 for Opus 5.5; index v5.0.1, a later revision than the rows above, Nov 9. Artificial Analysis · AA by effort level · AA on Opus 5.5 · The Decoder
Business adoption by provider
Share of US businesses on Ramp paying each company, by month (Ramp AI Index, via FRED). Anthropic’s surge came before Fable; neither Fable 5 nor Fable 5.1 moved it much.
Access, price and revenue, in order
- Fable 5 withdrawn for all users three days after launch to comply with US Commerce Department export controls; the controls were lifted June 30 and access returned from July 1. Anthropic · Al Jazeera
- On subscriptions, Fable 5 came back rationed: up to half of weekly usage limits on Max and Team Premium from July 20, usage credits on Pro and Team Standard. Startup Fortune
- Anthropic requires 30-day data retention on Fable traffic for safety monitoring; Ramp's chief economist cited data-retention rules as one drag on business take-up. Anthropic · Financial Times
- Anthropic's annualized revenue reached $65B in July, up from $47B in May but short of the roughly $80B investors had projected; OpenAI's rose 35% in the quarter to date, to more than $40B, after GPT-5.6's July launch. Financial Times
- Fable 5.1 kept the $10 / $50 price but cut cached input to $0.25, a quarter of Fable 5's; Anthropic put typical savings at about 25%. Anthropic
- Opus 5.5, which Anthropic says matches Fable 5.1 on most tasks, launched at $4 / $20; GPT-6 Sol and Luna followed within the hour at about half GPT-5.6's prices. TechNews · Sébastien Dubois
- Fable 5.2 launched at $8 / $40, 20% below Fable 5.1 and the first cut to a Fable list price, with cached input at $0.20. A fast mode, up to 2.5 times faster, costs $16 / $80. Anthropic · Anthropic pricing
- In launch week Anthropic gave plans that include Fable one to three extra Fable resets, by length of subscription, usable until March 8, 2027. Pro and standard Team seats got Fable for one week only, up to half of weekly usage limits and with one reset, through November 16; after that Fable on those plans runs on usage credits. Its launch post tells users to start with Opus 5.5 for most coding and knowledge work. Anthropic · Anthropic launch offers
- Fable 5.2 can run under zero data retention for eligible customers: Enterprise Frontier Safeguards, in private beta with Bridgewater, Salesforce and Stripe, runs safety monitoring on prompts and responses kept in the customer’s own infrastructure. Other customers stay on 30-day retention. Anthropic
All tracked launches
| Launch | Listed | Price in / out | Best week, share of spend | Best week, share of its lab | Week 6 vs best | Days to $25M | Outcome |
|---|---|---|---|---|---|---|---|
| Claude Fable 5.2premium | Nov 9 ’26 | $8 / $40 | 7.4% | 17% | pending | — | Too early |
| Gemini 4 Argonflagship | Oct 13 ’26 | $2 / $10 | 2.6% | 54% | — | — | Stayed niche |
| Claude Opus 5.5flagship | Sep 22 ’26 | $4 / $20 | 33.4% | 71% | 82% | 17 | Became the default |
| GPT-6 Solflagship | Sep 22 ’26 | $2 / $10 | 3.7% | 11% | — | — | Stayed niche |
| GPT-6 Astrapremium | Sep 4 ’26 | $10 / $50 | 22.8% | 56% | 22% | 16 | Breakout, faded |
| Claude Fable 5.1premium | Sep 1 ’26 | $10 / $50 | 8.9% | 30% | — | ≈41 | Frontier update |
| Claude Opus 5flagship | Jul 24 ’26 | $5 / $25 | 22.5% | 50% | 62% | 23 | Became the default |
| GPT-5.6 Solflagship | Jul 9 ’26 | $5 / $30 at launch | 7.7% | 44% | — | 48 | Stayed niche |
| Claude Fable 5premium | Jun 9 ’26 | $10 / $50 | 8.0% | 16% | — | ≈55 | Frontier update |
| Claude Opus 4.8flagship | May 27 ’26 | $5 / $25 | 19.8% | 33% | 95% | 28 | Became the default |
| GPT-5.5flagship | Apr 24 ’26 | $5 / $30 | 9.3% | 68% | — | 63 | Stayed niche |
| Claude Opus 4.7flagship | Apr 16 ’26 | $5 / $25 | 27.3% | 48% | 135% | 31 | Became the default |
| Gemini 3.1 Proflagship | Feb 19 ’26 | $2 / $12 | 3.9% | 35% | — | — | Stayed niche |
| Claude Opus 4.6flagship | Feb 4 ’26 | $5 / $25 | 26.7% | 49% | 104% | 50 | Became the default |
| Claude Opus 4.5flagship | Nov 24 ’25 | $5 / $25 | 21.4% | 39% | 104% | — | Became the default |
| Gemini 3 Proflagship | Nov 18 ’25 | $2 / $12 | 10.1% | 49% | — | — | Stayed niche |
| GPT-5flagship | Aug 7 ’25 | $1.25 / $10 | 3.2% | 43% | — | — | Stayed niche |
| Claude Opus 4.1premium | Aug 5 ’25 | $15 / $75 | 11.2% | 16% | — | — | Frontier update |
| Grok 4flagship | Jul 9 ’25 | $3 / $15 | 3.4% | 98% | — | — | Stayed niche |
| Claude Opus 4premium | May 22 ’25 | $15 / $75 | 11.6% | 18% | — | — | Frontier update |
Outcomes use the same tests for every launch: for the flagships, clearing the bar and holding it reads as “became the default”, never clearing it as “stayed niche”. Day by day for every launch: frontier-launch-curves.csv.
Method
Tokens. OpenRouter’s rankings-daily dataset: the 50 biggest models each UTC day plus one row for everything else. It covers public traffic through OpenRouter only, which leans toward developers and coding agents; enterprise traffic that goes straight to Anthropic, OpenAI, Google or the clouds is not in it. A model outside the day’s top 50 used fewer tokens than the 50th; for those days we estimate its tokens from the days around it, interpolating between the nearest days it was in the top 50 (or carrying the last one forward for up to a week), never above the 50th model’s count, and zero once it has been out for longer with no return. About 58B tokens a day is the cutoff now.
Spend. Each day’s tokens priced at the model’s OpenRouter list price that day, assuming 10% are output and half of input is billed at cached rates, as on the OpenRouter tracker. Shares are of all estimated OpenRouter spend that day, or of the model’s own lab’s. Premium models are priced at or above $40 per million output tokens; their low token counts and high prices make spend, not tokens, the fair measure.
Lining up launches. Day 0 is the day a model was listed in OpenRouter’s catalog. That is usually a partial day, so the curves, weeks and tests start with day 1, the first full UTC day. Days count only while a model was available, so Fable 5’s clock stops for the 18 days it was withdrawn. Curves are trailing seven-day averages; the first week averages the days so far, so a launch-day burst shows at the start. Weeks are days 1 to 7, 8 to 14 and so on.
Tests. Pull: a week among the first four at 15% or more of all OpenRouter spend. Staying power: week 6 at 60% or more of that best week. Upgrade: a week among the first four at 40% or more of the lab’s own spend. Speed: week 2 at 10% or more. The pull bar sits above every premium launch except GPT-6 Astra and below every Opus release since 4.5; these are our thresholds, set on October 25, 2026, before Fable 5.2 shipped, and applied to it unchanged.
Fable 5.2. This page was built in October with an empty slot for the next Fable, which we expected to be called Fable 5.5. Fable 5.2 was listed on November 9 and took the slot: it joins every chart and table with its listing date as day 0, under the same tests. While a launch has never been in the daily top 50, the page shows a ceiling in place of an estimate: the 50th model’s tokens each day, at the launch’s list price. Any other newly listed model at $40 or more per million output tokens that reaches the daily top 50 joins the premium launches automatically.
Updates. The data is pulled and the page rebuilt twice a day with the other trackers. Reported figures from outside OpenRouter (benchmarks, business spending, prices and access) are added as they are published, each with its source.
Token data: OpenRouter (openrouter.ai/rankings), as of 2026-11-13 14:41 UTC, CC BY 4.0. Spend, shares, estimates and tests are ours. Not investment advice. See also the Anthropic & OpenAI ARR Forecast.
Sources
- Anthropic: Claude Fable · Sep 1, 2026
- Anthropic: Statement on Fable 5 and Mythos 5 access
- Al Jazeera: US lifts restrictions on Anthropic's powerful AI models Fable and Mythos · Jul 1, 2026
- Startup Fortune: Anthropic limits Claude Fable 5 access as it runs out of compute · Jul 18, 2026
- Financial Times: Anthropic's best AI model struggles to attract users as cheaper tools thrive · Aug 23, 2026
- Simon Willison: Ramp AI Index model shares, in a note on the FT report · Aug 23, 2026
- OfficeChai: These are the AI models with the most customer spend, as per Ramp · Sep 23, 2026
- Artificial Analysis: Claude Fable 5 and Mythos 5 on the Intelligence Index · Jun 9, 2026
- Artificial Analysis: Benchmarking GPT-6 Astra · Sep 9, 2026
- OfficeChai: Claude Opus 5.5 jumps to top spot on Artificial Analysis Intelligence Index · Sep 22, 2026
- Artificial Analysis: Gemini 4 Argon vs GPT-6 Astra · Oct 24, 2026
- TechNews: Anthropic upgrades Claude with new Opus 5.5 model · Sep 23, 2026
- Sébastien Dubois: GPT-6 Sol and Luna: same smarts as GPT-5.6, half the price · Sep 22, 2026
- Ramp AI Index via FRED: AI adoption by Ramp businesses: Anthropic · Oct 1, 2026
- Ramp AI Index via FRED: AI adoption by Ramp businesses: OpenAI · Oct 1, 2026
- Anthropic: Introducing Claude Fable 5.2 and Claude Mythos 5.2 · Nov 9, 2026
- Anthropic: Claude Fable 5.2: pricing and availability · Nov 9, 2026
- Anthropic: Claude Fable 5.2 launch offers · Nov 9, 2026
- Artificial Analysis: Claude Fable 5.2 (max): intelligence, performance and price · Nov 9, 2026
- Artificial Analysis: Claude Fable 5.2 vs Gemini 4 Carbon, by effort level · Nov 9, 2026
- Artificial Analysis: Claude Opus 5.5 (max) vs Claude Fable 5.1 (high) on index v5.0.1 · Nov 9, 2026
- The Decoder: Claude Fable 5.2 and Mythos 5.2 cement Anthropic's AI lead just before its IPO · Nov 9, 2026