Fable 5.2 Just Dropped Full Breakdown (No Hype)
🚀 Final Phase: https://www.skool.com/finalphaseai/about 🆓 All my free skills (free Skool): https://www.skool.com/nextphase-5790 Claude Fable 5.2 just came out, and the price matters most. It's 20% cheaper than Fable 5.1, but still 2× Opus 5.5 per token. Per task, though, it can come out cheaper, because it uses about half the tokens. I compared it to the 3 models you're actually choosing between: Opus 5.5, GPT-6 Astra (OpenAI) and Gemini 4 Carbon (Google). 💸 The price, in plain words • Fable 5.1: $10 in / $50 out per million tokens. Fable 5.2: $8 / $40. Opus 5.5: $4 / $20. • What you really pay is the tokens it takes to finish the task. On Vals AI (the first independent test), Fable used about half the tokens Opus did: about $18 vs $25 for the same set of tests, done in 44 minutes vs almost an hour and a half. • On Anthropic's own coding test, Fable on High scores the same as Opus on Max for about half the money (about $7 vs $13 a task). 👨💻 Coding Anthropic's test: real open-source projects, then would the people who run them accept the code? Fable 5.2 about 49%, Opus about 45%, Astra and Carbon a bit behind. Fable 5.1 was at 32%, so it went from last to first. Which effort? Extra High gets that top score but costs about $14 a task (Opus on Max: about $13). High gets you Opus-level code for about $7, that's the sweet spot. Max scores LOWER than Extra High and costs about $45 a task, so don't use it. On Vals: #1 of 127 models for building full apps from a prompt, Opus wins on fixing bugs that come with screenshots, and they're tied in the terminal. 🤖 Agents Real freelance projects (Remote Labor Index): Fable on High about 41%, Opus and Astra about 21%. The biggest gap in the whole launch. Just using a computer (OSWorld): all close, Fable first, Opus right behind. 💼 Business work Business workflows (AutomationBench): Gemini 4 Carbon is #1, Fable 2nd, Opus and Astra a lot lower. Office work (AA-Briefcase): Fable and Opus basically tie, Opus wins 1 of the 2 scores. Reading charts and graphs: Fable, Carbon and Astra super close, Opus is the weak one. 📊 The independent test (Vals AI Index) Real-work tests like finance, coding, legal and tax. Fable 5.2 is #1 of 36 models, but by less than half a point: Sonnet 5.5 2nd, Carbon 3rd, Opus 4th, all basically tied. Astra is clearly behind. And on tax, Fable is #27. The top 4 are neck and neck, Fable just gets there faster and cheaper than Opus. 🏁 Model by model • Fable 5.2: the hardest, biggest jobs (long coding projects, agents running for hours, research). Use High, not Max. • Opus 5.5: still the everyday model. Half the price per token, almost the same coding. Anthropic says it themselves: start with Opus, go up to Fable when the job is really hard. • Gemini 4 Carbon: best for business automations and the cheapest of the 4 on Vals. But you can't use it yet: only a small group of cybersecurity teams, no date. • GPT-6 Astra: doesn't win anything here. The fastest on Vals, but behind on most tests, and it cost more than Fable on the Vals run. 🔑 How to get it All paid Claude plans, in the apps and in Claude Code (type /model and pick it). It starts on Medium effort, so put it on High for the sweet spot. Launch promo this week: on Pro you can use up to half your weekly limits on Fable, and plans that already include Fable get 1 to 3 free resets depending on how long you've been subscribed. New fast mode: up to 2.5× faster, but double the price ($16 / $80). 🧠 My take The first Fable that makes sense for normal people. Big builds, like a whole new editor skill: Fable on High. Everyday stuff (editing my videos, small fixes): Opus for now, until I test them side by side myself. Most launch-day numbers are Anthropic's own, so I'll make a follow-up when Artificial Analysis and Arena post their results. 📚 Sources (checked on launch day, Nov 9 2026) Anthropic launch page: https://www.anthropic.com/claude-fable-and-mythos-5-2 Vals AI Index: https://www.vals.ai/benchmarks/vals_index Vals AI, Fable 5.2: https://www.vals.ai/models/anthropic_claude-fable-5-2 ✏️ Correction At 1:10 I say Gemini 4 Carbon "just got released". It's announced, but it's still not public: only a small group of cybersecurity teams can use it. ⏱️ Chapters 0:00 The price: per token vs per task 1:27 Where these numbers come from 1:44 Coding: who wins 2:16 Which effort to use (not Max) 2:48 Independent coding tests 3:10 Agents: the biggest gap 3:50 Business work: Carbon wins 4:18 The independent test (Vals AI) 4:55 Model by model 5:47 How to get Fable 5.2 6:19 My take 💬 Are you switching to Fable 5.2, staying on Opus, or using a different model? Tell me in the comments. 🔔 Subscribe, I drop free skills every week. #ClaudeFable #Fable52 #Claude #Anthropic #ClaudeCode #ClaudeAI #Opus55 #GPT6 #Gemini4 #AIModels #AIBenchmarks #AINews #AICoding #AIAgents #LLM
