Open Chat
Comparison

Ox Alpha vs Fable 5

Head-to-head on an independent benchmark of 10 real-world coding tasks: the stealth model against Anthropic's Fable 5, max effort.

80%
Ox Alpha
ox-alpha · 8/10 tasks
VS
65%
Fable 5
fable-5 [max] · mean pass rate
Task by task

Where each model wins.

5
tasks where Ox Alpha had the edge
1
tasks where Fable 5 had the edge
3
tasks both solved cleanly
1
tasks neither solved
Taskfable-5 [max]ox-alphaVerdict
anko-typed-variable-bindings 4/4 Both solved
arktype-json-schema-refs 2/4 Ox Alpha
fastapi-deprecation-headers 4/4 Both solved
helm-unified-manifest-stream 4/4 Both solved
igel-persist-feature-schema 3/4 Ox Alpha
katex-multicolumn-array-spans 2/4 Ox Alpha
meriyah-explicit-resource-decl 1/4 Ox Alpha
query-persist-restored-state 2/4 Ox Alpha
scc-bounded-memory-spilling 4/4 Fable 5
vulture-persistent-analysis-cache 0/4 Neither
Mean on these 10 65% 80%

Independent community benchmark — 10 real-world coding tasks. Reference models were scored as passes out of 4 attempts per task; Ox Alpha was recorded as pass/fail. "Edge" = Ox Alpha passed while Fable 5 missed at least one attempt, or vice-versa. Third-party data presented as published — 10 tasks is a small sample, so read it as directional.

What the numbers say

Ox Alpha vs Fable 5, max effort
  • 8 of 10 solved by Ox Alpha, versus 4 tasks that Fable 5 passed on all four attempts.
  • Fable 5's strongest ground: 1 task it passed that Ox Alpha missed — scc-bounded-memory-spilling.
  • Both stumbled on 1 task — a reminder that no model is a guaranteed pass on real-world bugs.

Ox Alpha at a glance

What's known about the stealth model
Type
Reasoning model
Context window
1,048,576 tokens
Max output
131,072 tokens
Inputs
Text · Image · Video
Developer features
Tool calling · JSON output
Creator
Unknown (stealth)

About Fable 5

Anthropic · benchmarked build: fable-5 [max]
  • Anthropic's flagship Claude Fable 5, run at maximum reasoning effort — the strongest reference model in this benchmark.
  • Mean pass rate on these 10 tasks: 65% — passing at least one attempt on 9 of 10.

Why it matters

Reading a 10-task benchmark honestly
  • These are real repository tasks, not toy puzzles — the kind of work a reasoning model is built for.
  • Ten tasks is a small sample; a few tasks either way would move the means. Use it to decide what to try, not to declare a winner.
  • The best benchmark is your own problem. Ox Alpha is free to try right now.

Run your own benchmark.

Paste the bug, the spec, or the whole repo. Ox Alpha is free while it's in stealth.

Try Ox Alpha