Anthropic's leaderboard winner takes 11% of Anthropic's own platform spend
Enterprises buy the cheapest model that clears their bar. On Ramp's July billing data, that leaves Anthropic's flagship with about an eighth of its maker's platform spend.
The Engineer · Build desk
What happened
- Financial Times reporting cited by dev.to puts Anthropic's top model, Fable 5, at 11 percent of enterprise spending on Anthropic's own platform.
- Ramp's July 2026 index shows older Opus 4.8 at 28.0 percent of tracked spend against Fable 5 at 8.0 percent, with eight other Anthropic models below.
- Fable 5 leads SWE-Bench Pro at 80.3 percent, eleven points clear of the nearest competing model.
- The Ramp index is built from billing data covering 70,000 companies that use Ramp corporate cards, not surveys or self-reported usage.
Why it matters
- What it limits constraint Rank has stopped setting price. A model can lead its category and still lose the mix inside its own vendor's platform, which caps what any provider can charge for being first on a leaderboard.
- What doesn't add up contradiction If the 4-series is holding rather than handing spend to a cheaper 5-series, the story is buyers refusing to migrate at all, which is a harder problem for a vendor than steering them down-market.
- Who's exposed exposure OpenAI's undercut of Opus 5 runs on promotional pricing through November, so anyone who reprices their routing around it inherits a renewal cliff they did not negotiate.
Next story · Invest · Sep 22
Sol's 27-cent benchmark task undercuts Opus 5 by more than eleven timesConfirmed16 publishers
Ten Anthropic models in Ramp's July table total 63.0 percent of tracked spend [11], which means those shares are slices of something wider than Anthropic, and no statement about Anthropic's internal mix survives without renormalising them. Do that and Fable 5 comes to 12.7 percent of the Anthropic spend Ramp can see [10], close enough to the 11 percent in the FT's reporting [1] that the two are probably measuring the same thing. The rest of the renormalised table is the part a finance team should be shown: 4-series models plus Haiku 4.5 hold 76.0 percent, and everything badged 5 holds 24.0 percent [12][13]. Three quarters of the money is sitting on last generation.
The mechanism is arithmetic. Fable 5's list price is twice Opus 4.8's on both input and output [14], so it has to return twice the value per task, and the dev.to writeup of the FT data puts the real gain at roughly 1.1x on the workloads it names, summarisation and document extraction among them [22]. That is the whole story of the 11 percent.
Where the two accounts split is Opus 5. The dev.to piece has it in the mid-teens and overtaking Fable within weeks of its July 24 launch [16]. Ramp's July table has it at 3.5 percent against Fable's 8.0 [6], which renormalises to 5.6 percent of Anthropic spend [18]. Simon Willison supplies the reason for the gap himself: July contained one week of Opus 5 [23]. So the cannibalisation claim is a forecast rather than a reading, and CNBC's report that Opus 5 beats Fable 5 on coding and knowledge-work evaluations at half the price [25] is the case for expecting it to come true, not evidence that it has.
One number in the dev.to piece does not survive a check. It calls the gap between Fable 5 output and DeepSeek V4-Flash output 71x [17]; the two prices it prints, $50 and $0.28 per million tokens [5][8], are 179x apart [19]. The 71x figure is what you get from the discounted $20 output price on OpenAI's Sol measured against the same floor [7][20]. The direction of travel is the same either way. A price table that mislabels its own ratio by a factor of two and a half [21] is still not the slide to paste into a procurement deck.
None of this is a demand problem. Anthropic's annualized revenue reached $65 billion in July, up from $47 billion in May, with 6,000 accounts spending $100,000 a year or more [3]. It is a mix problem, and on the buyer side mix is usually one config value. The default model in a gateway gets set once, by whoever wired it up, and on these prices that line carries a factor of two on every token that passes through it. It is rarely owned by anyone who reads the invoice.
What to watch
- Ramp's August index, the first full month with Opus 5 in it, and whether the 4-series share falls or holds.
- Whether OpenAI's Sol discount expires in November or becomes the list price.
- Anthropic's Q3, which it has told investors it expects to be profitable, and whether the mix shows up in margin rather than revenue.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence55
- Adoption25
- Hype gap+30
- Incentives40
- Confidence55
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
According to data published by the Financial Times, Fable 5 accounts for just 11 percent of enterprise spending on Anthropic's tools, measured as a share of spend on Anthropic's own platform.
- [2]
The Ramp AI index uses billing data from 70,000 companies that use Ramp credit cards to estimate model adoption.
- [3]
Anthropic's annualized revenue for July reached $65bn, up from $47bn in May, and the company told investors it had 6,000 customers spending $100,000 annually or more.
- [4]
Fable 5 tops SWE-Bench Pro at 80.3 percent, eleven points ahead of its nearest competitor.
- [5]
Fable 5 costs $10 per million input tokens and $50 per million output tokens, exactly double Opus 4.8 and Opus 5; Opus 4.8 is priced at $5/$25 per million tokens.
- [6]
Ramp's breakdown of Anthropic model spend for July 2026: Opus 4.8 28.0%, Sonnet 4.6 8.3%, Fable 5 8.0%, Opus 4.6 6.9%, Sonnet 5 3.6%, Opus 5 3.5%, Opus 4.7 1.7%, Sonnet 4.5 1.3%, Haiku 4.5 1.0%, Opus 4.5 0.7%.
- [7]
On August 21 OpenAI cut GPT-5.6 Sol prices by over 20 percent, input from $5 to $4 and output from $30 to $20 per million tokens; Sol now undercuts Claude Opus 5 on both, and the cut is promotional through November.
- [8]
DeepSeek V4-Flash runs at $0.14 per million input tokens and $0.28 per million output tokens.
- [9]
Anthropic expects Q3 to be profitable according to the same model it used to declare Q2 profitable.
- [10]
Renormalised to the Anthropic models Ramp lists, Fable 5 is 12.7 percent of Anthropic spend.
- [11]
The ten Anthropic models in Ramp's July table sum to 63.0 percent of tracked spend, so the shares are not shares of Anthropic-only spend.
- [12]
The 4-series models plus Haiku 4.5 total 47.9 points, which is 76.0 percent of Ramp's listed Anthropic spend.
- [13]
The 5-generation models (Fable 5, Sonnet 5, Opus 5) total 15.1 points, which is 24.0 percent of Ramp's listed Anthropic spend.
- [14]
Fable 5 is priced at exactly 2.0x Opus 4.8 on both input and output tokens.
- [15]
dev.to states that Opus 4.8 surged from 15 percent of Anthropic model spend in early June to over 50 percent by late July.
- [16]
dev.to states that Opus 5, released July 24 at half Fable 5's price, overtook Fable within weeks and sits at a mid-teens percentage of Anthropic model spend.
- [17]
dev.to states there is a 71x spread between Fable 5 output pricing and DeepSeek V4-Flash output pricing.
- [18]
Renormalised, Opus 5 is 5.6 percent of Ramp's listed Anthropic spend in July.
- [19]
The two output prices dev.to prints, $50 for Fable 5 and $0.28 for DeepSeek V4-Flash, are about 179x apart, not 71x.
- [20]
The discounted $20 output price for GPT-5.6 Sol against DeepSeek V4-Flash output is 71.4x, matching the ratio dev.to labels as Fable-to-DeepSeek.
- [21]
The stated 71x spread understates the ratio implied by the quoted prices by a factor of about 2.5.
- [22]
dev.to argues that at double the price Fable 5 needs to deliver 2x the value per task but for most enterprise workflows delivers maybe 1.1x, naming summarization, classification, code review and document extraction as the typical workloads.
- [23]
Simon Willison writes that the July figures look reasonable given that Opus 5 was only released on July 24th, and that they support the idea that Fable's cost has made it a less popular model.
- [24]
dev.to concludes that enterprises are not choosing the smartest model but the cheapest model that clears their quality bar.
- [25]
CNBC reported that Opus 5 outperforms Fable 5 on coding and knowledge work evaluations while costing 50 percent less.
Sources
2 independent publishers whose own reporting we read for this story.
- dev.toAnthropic's Best Model Can't Hold Users
1 article · August 24, 2026
- simonwillison.netAnthropic’s best AI model struggles to attract users as cheaper tools thrive
1 article · August 23, 2026
Topics and entities
Topics
- Enterprise AI Adoption
- Model pricing and inference economics
- AI company finances