Ox Alpha Vs Fable 5
You've seen the hype for both, but you still have to answer one practical question: which model should actually power your next coding or agent build — the free "stealth" model everyone on developer forums is trying to identify, or Anthropic's newest frontier flagship?
Ox Alpha appeared on OpenRouter on August 20, 2026 as a free reasoning model built for "coding, sustained agentic work, and production workloads," and within days it had set off a guessing game about who built it. Two months earlier, on June 9, Anthropic launched Claude Fable 5 — a Mythos-class model it calls its most capable model ever made generally available. They are the two most talked-about coding and agentic models of the late-summer 2026 cycle, and the Ox Alpha vs Fable 5 decision keeps coming up in the same threads.
This comparison is written from the official OpenRouter listing for Ox Alpha, Anthropic's launch announcement and Fable 5 product page, and TechCrunch's reporting on the Ox Alpha mystery. Where the identity of Ox Alpha's developer is unconfirmed, we say exactly that — we do not guess.
Read to the end and you'll be able to make the call in a few minutes: price, context, modalities, tool calling, and the risk of depending on an anonymous provider — plus a free third option you can test today.
TL;DR
Ox Alpha is a free, 1M-context stealth coding model on OpenRouter whose developer chose to stay anonymous; Fable 5 is Anthropic's $10/$50-per-million-token frontier flagship with safety classifiers, enterprise support, and a known vendor. If cost is everything and you can tolerate an unverified provider, Ox Alpha is worth a look. If you need a proven, supported model for long-horizon production work, Fable 5 is the safer pick. And if you want near-frontier performance without either trade-off, test GLM 5 free at glm5.app before you commit.
What This Article Solves
Most coverage of Ox Alpha is mystery-shopping, and most coverage of Fable 5 is a press release. This article exists to give you the one comparison that resolves the actual buying decision for a coding or agent team: what each model verifiably offers, what is still unknown, and which one to route to for which workload. The section most other posts skip — the real risk of depending on an anonymous stealth provider — is treated as a first-class decision factor here, alongside the numbers.
Ox Alpha vs Fable 5: At a Glance
| Dimension | Ox Alpha | Claude Fable 5 |
|---|---|---|
| Developer | Anonymous third party ("stealth") | Anthropic |
| Released | Aug 20, 2026 | Jun 9, 2026 (restored Jul 1, 2026) |
| Price | Free | $10 / $50 per million tokens (in/out) |
| Context window | 1,048,576 tokens (1M) | Not publicly disclosed |
| Max output | 131,072 tokens | Not publicly disclosed |
| Input / output | Text, image, video → text | Text and vision (multimodal) |
| Tool calling | Yes (tools, tool_choice, JSON response_format) | Yes, mature agentic tooling |
| Primary use case | Coding, agentic work, production workloads | Long-horizon coding, knowledge work, research |
| Safety | No published safety classifier program | Classifiers on cyber, bio/chem, distillation |
| Enterprise support | None — single anonymous provider | Plans, API, AWS / Google Cloud / Foundry |
What Is Ox Alpha?
Ox Alpha is a reasoning model served through OpenRouter as a "stealth" listing. The OpenRouter page states that it is "developed and operated by a third-party provider who has chosen to remain anonymous during this preview," and that OpenRouter only routes requests — it is not the developer, owner, or provider. The listing is explicit that prompts and completions are retained by the provider and are not used for training.
The specs are real and checkable: a 1,048,576-token context window, up to 131,072 output tokens, text/image/video input with text output, and support for tools and tool_choice function calling plus response_format for JSON output. It is currently free, hosted by a single provider, with roughly 26 tokens/second throughput and about 99.4% three-day availability reported by OpenRouter.
Who actually built it is the open question. TechCrunch reported that speculation first centered on the GLM models from Chinese lab Z.ai, then on an unreleased Microsoft MAI model, and that "people seem less sure of anything" within days. Stripe CEO Patrick Collison called it "very impressive" on X — and Stripe is in the process of acquiring OpenRouter. No one has publicly confirmed Ox Alpha's developer. Treat every identity claim as unverified until a named party confirms it.
What Is Claude Fable 5?
Claude Fable 5 is Anthropic's first generally available Mythos-class model — a tier above its Opus line. Per Anthropic, it is state-of-the-art on nearly all tested benchmarks, with exceptional performance in software engineering, knowledge work, vision, and scientific research, and its lead grows with task length and complexity.
The numbers that matter for a decision: $10 per million input tokens and $50 per million output tokens, with the existing 90% input-token discount for prompt caching — less than half the price of Anthropic's earlier Mythos Preview. It is available on Pro, Max, Team, and Enterprise plans, via the Claude API under the model name claude-fable-5, and through AWS, Google Cloud, and Microsoft Foundry.
Two caveats are worth knowing. First, Fable 5 ships safety classifiers for cybersecurity, biology and chemistry, and distillation: flagged requests fall back to a less capable Claude model instead of a refusal. Anthropic says more than 95% of sessions trigger no fallback, and you are not charged Fable prices for rerouted requests — but the classifiers can occasionally catch harmless requests. Second, Anthropic requires 30-day data retention on Mythos-class traffic for safety monitoring, with deletion in almost all cases after that window. Availability was also not smooth: access was suspended on June 12 and restored on July 1, 2026.
Head to Head: The Decision Factors
Price. This is the widest gap. Ox Alpha is free with a 1M context; Fable 5 costs $10/$50 per million tokens (before the 90% prompt-cache discount). For a high-volume agent workload, that is the difference between a rounding error and a real line item. Rule of thumb: if your unit economics are cost-sensitive and you are still evaluating models, test the free one first.
Context and long-horizon work. Ox Alpha's 1M context is officially documented and large enough for whole-repository prompts. Anthropic does not publicly disclose Fable 5's context window, though its marketing centers on long-running, multi-day autonomous work — the "longer the task, the larger the lead" claim. If you must architect around a documented ceiling, only Ox Alpha gives you one in writing.
Modalities. Ox Alpha accepts text, image, and video input; Fable 5 is documented as strong on vision (figures, charts, screenshots) within Anthropic's multimodal stack. For workflows that combine text with visual context, Ox Alpha's listed video input is a concrete spec, while Fable 5's vision strength is benchmark- and use-case-based.
Tool calling and agents. Both support function calling. Anthropic's tooling ecosystem — Claude Code, agent harnesses, enterprise fallback APIs — is the most mature in the industry and is the practical reason many teams pick Fable 5. Ox Alpha offers OpenAI-compatible tool calling and JSON output through OpenRouter, with no vendor tooling around it.
Trust and risk. This is where the two diverge most. Fable 5 has a named vendor, a published system card, transparent safety classifiers, enterprise contracts, and a documented (if bumpy) launch. Ox Alpha has an anonymous operator, a single provider, no published safety program, and no commercial terms beyond OpenRouter's stealth-model terms. If a model sits in your production critical path, "we don't know who runs it" is a real cost.
Decision Framework: Which One Should You Pick?
| Your situation | Pick |
|---|---|
| Experimenting, low stakes, cost-sensitive evaluation | Ox Alpha — free, 1M context, worth testing before you spend |
| Production coding or agent loops that must not break | Fable 5 — named vendor, support, enterprise fallback APIs |
| You need a documented 1M context window in writing | Ox Alpha — 1,048,576 tokens confirmed on the listing |
| Hardest, longest, most ambiguous problems | Fable 5 — Anthropic's claimed state-of-the-art on long tasks |
| Compliance, data residency, or vendor audit requirements | Fable 5 — known operator, documented retention; Ox Alpha is a non-starter |
| Near-frontier reasoning without either trade-off | GLM 5 — see below |
The common-sense pattern: route experimental and cost-sensitive volume to a free or cheap model, escalate the gnarly long-horizon work to a proven flagship, and keep a fallback path for both.
The Third Option: GLM 5
Before you commit to an anonymous provider or a premium flagship, consider the middle path. GLM 5 — Zhipu AI's fifth-generation frontier model, available free at glm5.app — pairs open-source, frontier-class reasoning and coding (745B total parameters, MoE with ~44B active, 200K context) with an actively maintained hosted platform. You get a named vendor, no API key or GPU cluster, and a real coding-and-agentic workload you can judge for yourself in the browser.
If you are weighing Ox Alpha vs Fable 5 for a build, spend ten minutes running your actual prompts on GLM 5 at glm5.app first. It is the fastest way to calibrate what "close enough to frontier" feels like before you lock in either architecture — and it is free to start.
Frequently Asked Questions
Is Ox Alpha free? Yes. The OpenRouter listing prices Ox Alpha at zero for both prompt and completion tokens during this preview period.
Who made Ox Alpha? Unconfirmed. OpenRouter lists it as a "stealth model" from an anonymous third-party provider. TechCrunch reported speculation ranging from Z.ai's GLM team to an unreleased Microsoft MAI model, with no named confirmation.
Does Fable 5 really cost $10/$50 per million tokens? Yes. Anthropic's launch announcement and Fable 5 product page both state $10 per million input and $50 per million output tokens, with a 90% input discount for prompt caching. US-only inference is available at 1.1x pricing.
Which model is better for coding? For long-horizon, high-stakes, autonomous engineering, Fable 5 is the proven flagship with the mature tooling ecosystem. For cost-sensitive volume and quick evaluation, Ox Alpha's free 1M-context access is the lower-risk experiment — assuming you accept an unverified provider.
Bottom Line
Ox Alpha vs Fable 5 is not a "better model" debate; it is a risk-allocation decision. Ox Alpha is a free, 1M-context coding and agentic model with real specs and a genuinely unknown operator — fine for evaluation, risky as a production dependency. Fable 5 is Anthropic's expensive, documented, safety-wrapped frontier flagship with the strongest agent tooling in the industry — the safe pick for work that must not break. Either way, run your own workload before you choose.
Start with the free option that has a named vendor: test GLM 5 at glm5.app and see how it handles your real prompts — reasoning, coding, and agentic tasks in one place, no API key required.
By the GLM 5 Team — we build the free gateway to GLM 5 at glm5.app, and we write comparisons from official documentation and reporting, not vendor press releases. Figures reflect official sources as of August 24, 2026; pricing and availability change frequently, so verify against the official pages below before budgeting.
Sources
- OpenRouter — Ox Alpha listing — Official specs: free pricing, 1M context, 131K output, modalities, tool calling, stealth terms, release date
- Anthropic — Claude Fable 5 and Claude Mythos 5 announcement — Official launch, pricing, safety classifiers, availability timeline
- Anthropic — Claude Fable 5 product page — Official availability, pricing, safeguards, and data-retention details
- TechCrunch — Who's behind the new 'stealth model' Ox Alpha? — Reporting on the Ox Alpha identity question, Collison quote, and Stripe–OpenRouter context
Scope note: Ox Alpha facts (release date, context, output limit, modalities, tool calling, single-provider hosting) come from the OpenRouter listing; its developer's identity is unverified, and all identity claims are attributed to TechCrunch's reporting. Fable 5 facts (pricing, capabilities, safeguards, retention, restoration dates) come from Anthropic's announcement and product page. Anthropic does not publicly disclose Fable 5's context window. GLM 5 details are from glm5.app. Verify current pricing and availability on the official pages above.
