Anthropic's best AI model struggles to attract users as cheaper tools thrive
Anthropic’s top-end Fable 5 model is drawing relatively little use compared to cheaper options like Opus, GPT‑5.6 “Sol,” and fast-rising Chinese open-weight models, despite often being seen as more capable for complex, long-horizon coding and research tasks. Commenters point to high costs, aggressive usage limits, intrusive safety guardrails, lack of zero-data-retention for enterprises, and erratic pricing and policy changes as major deterrents, especially when “good enough” models are now plentiful and far cheaper. Many see this as evidence that intelligence gains beyond a certain point bring diminishing practical returns, pushing AI toward a commoditized, cost-sensitive market rather than a runaway premium frontier.
Model Adoption and “Good Enough” Intelligence
- Many commenters say Fable 5’s frontier-level capability isn’t needed for most day‑to‑day work (coding, email, summaries).
- Cheaper models (Opus 4.6/4.8, Sonnet, GPT 5.6 Sol, DeepSeek, GLM, Qwen, Kimi) are seen as “good enough,” especially when paired with good harnesses/agents.
- Some argue we may be near a “good enough” plateau where incremental intelligence adds little value outside frontier research or very complex engineering.
Pricing, Quotas, and Business Strategy
- Strong frustration with Anthropic’s shifting subscription tiers, Fable access windows, and opaque 5‑hour / weekly limits.
- Several users report hitting limits mid‑session and feeling forced to “token-maxx” or buy multiple accounts or higher tiers.
- Many perceive Anthropic as aggressively monetizing tokens and acting like a “token merchant,” in contrast to competitors’ more generous or predictable quotas.
Quality, Regressions, and Guardrails
- Widespread sentiment that Opus 5 is slower, more verbose, more condescending, and often worse than Opus 4.6/4.8; some suspect silent downgrades or infrastructure changes.
- Fable is praised by a minority for long‑horizon planning, complex refactors, and autonomous multi‑hour code transformations, but others find Sol or K3 comparable or better at lower cost.
- Safety guardrails are a major pain point: Fable and Opus frequently refuse benign security, biology, auth, or systems questions, or silently downgrade models.
Enterprise and Compliance (ZDR)
- A key blocker for Fable in enterprises is lack of zero data retention; multiple organizations forbid its use for this reason.
- Some see this driven by regulatory pressure and Anthropic’s desire to analyze multi‑request attack patterns. Others view it as surveillance and a trust issue.
Privacy, Trust, and UX
- The prompt‑fingerprinting / steganography incident is cited as a serious privacy violation and trust breaker.
- Users dislike A/B‑tested behavior, auto “auto‑mode” toggles, and changing model personalities; they want stable, predictable tools.
Competition, Local Models, and Bubble Talk
- Many report migrating heavy work to OpenAI Codex/Sol, Chinese models via OpenRouter, or self‑hosted Qwen/GLM/DeepSeek, often purely on cost and reliability.
- Several predict LLMs becoming cheap commodities run locally; foundation labs’ moats and AI‑stock valuations are viewed skeptically.