Model profitability analysis: Opus 5.5 beats Fable 5.1 at half the price, Grok loses on every task

Wsz2020 · reddit · 2026-09-23

Using ArtificialAnalysis.ai's intelligence score vs. output tokens-per-task chart, the author maps which models are actually profitable:

• The bottom-right quadrant (low scores, high token burn) forces deep discounts: Grok and Muse fit this pattern — the more people use them, the more Musk and Zuck lose.
• Counterintuitively, the best model isn't the priciest: brand-new Opus 5.5 beats Fable 5.1 at less than half the price, cutting deeply into Anthropic's margins.
• Fable 5.1 had been the most profitable, most expensive, and best-performing model — until today. The author predicts a Fable 5.5 from Anthropic as soon as engineers can ship it.

Takeaway: the mismatch between pricing and capability/cost determines whether a vendor profits more with every query or loses money.

Original post →

More from Venture

• Robotics data startup Eidon.AI shuts down after 2+ years: right thesis, brutal business — xiao_ted · 2026-09-23
• Founder says Ohio VC balked at 4% equity in $50M pre-revenue $2M AI seed round — MickeySteamboat · 2026-09-23
• AI-era investing: passing on solo founders makes less sense as they hit big ARR — julianweisser · 2026-09-23
• Cheaper Inference launches OpenAI-compatible API reselling models at up to 60% off — VeryWellVersed · 2026-09-23
• BlackRock paper argues broad AI adoption may drive new demand for digital assets — learnedall · 2026-09-23
• Garry Tan: YC has backed over a dozen data and RL environment startups doing $10M+/yr — garrytan · 2026-09-23

Venture channel →