MANIFOLD
Was sonnet 5 meant to be released instead of Opus 4.6?
17
Ṁ100Ṁ1k
Jun 30
3%
chance

Resolution criteria

This market resolves YES if credible evidence emerges that Claude Sonnet 5 was planned for release on the week starting February 2nd, but later decided to release Opus 4.6 after failed deployment attempts. NO otherwise by close date

Background

Anthropic released Opus 4.6 on February 5, 2026. Prior to the release, there were persistent rumors of a Sonnet 5 release, with prediction markets hovering around 80% odds. Opus 4.6 is priced 66% higher than Sonnet, creating an incentive to brand a capable, but cheap to run, model as Opus. However, there haven't been any leaks calling the released model Sonnet, and multiple organizations with early access all announced Opus 4.6 upon release.

Considerations

Claude itself suggested there may be some relationship between the internal codename "Fennec" (associated with Sonnet 5 leaks) and what became Opus 4.6, but acknowledged the full story is likely more complicated than a simple rebrand, and that Sonnet 5 might still release separately.

I will not bet on this market

Market context
Get
Ṁ1,000
to start trading!
Sort by:

I am surprised this is so low, I thought there was a good chance this happened. I guess we likely won't have evidence or credible report on this though which will make this resolve NO

bought Ṁ10 YES

@Mochi ok i put some money up on this because there's no way they made Opus 4.6 2.5 times faster w/o just renaming it sonnet???

@Bayesian would love to hear your theories on this though

@prismatic yeah that seems like pretty decent evidence

@prismatic strongly disagree it's 6x more expensive. clearly just beefier serving infrastructure.

@sylicon they price it whatever they can het away with but tbf i dont think they got the balls to surcharge 50x for sonnet 5 and to slow down normal use

that is a next level conspiracy

@Bayesian yeah I'm also really skeptical that better infra leads to a 2.5x increase in speed. OpenAI overhauled its entire stack for a 40% speedup and there are no signs this model is being served on bleeding edge infra like cerebras/blackwell.

@prismatic

you can trade gpu throughput for per-user throughput. and more aggressive speculative decoding and stuff like that

@Bayesian good stuff, forgot abt how powerful spec decoding could get

opened aṀ25YES at 9% order

"I will not bet on this market"
But isn't your twin brother betting?

@adonisds lol good catch, I can’t control all markets he bets, will make him sell as soon as I find out! I promise it won’t affect my decisions here even if he holds positions

© Manifold Markets, Inc.TermsPrivacy