Skip to main content
MANIFOLD
When will we have an open-weights model on-par with Mythos/Fable
9
Ṁ100Ṁ853
2031
December 3, 2026
86%
Before 2027
98%
Before 2028
98%
Before 2029
98%
Before 2030

"on-par" is defined as within 3 points of Fable/Mythos on benchmarks on average. If Fable scores lower on a benchmark compared to Mythos, I will use Mythos's score

The model should match Mythos on its ability to find vulnerabilities, too.

Market context
AI
Technical AI Timelines
OpenAI
AI model releases
Fable
Get
Ṁ1,000
to start trading!

People are also trading

Will OpenAI release a Mythos-class model before Fable is reenabled?
1% chance
Will Anthropic release an open-weights model in 2026?
10% chance
Weights in largest open-weight AI model before October 2026?
Weights in largest open-weight AI model before 2027?
When will Llama 4 Behemoth weights become publicly available?
10/14/26
Will a non-Anthropic company make a Mythos-class model generally available worldwide in 2026?
86% chance
An open weight model beats Opus 4.7 by EOY26
86% chance
Will "Mythos-class" model be called Mythos?
11% chance
Will GigaChat release an open-weights model with ≥100B parameters by the end of 2026?
45% chance
Will Anthropic require training-consent for any frontier model by mid-2027?
16% chance
Sort by:

@traders Kimi k3 does not match Mythos in finding vulnerabilties

https://www.nist.gov/news-events/news/2026/07/uk-aisi-caisi-preliminary-assessment-kimi-k3s-cyber-capabilities

In fact, it falls behind GPT5.5 and Opus 4.8

https://semgrep.dev/blog/2026/kimi-k3s-code-security-results-lack-precision/#our-results

Kimi K3 Benchmarks: Strong on Paper, Weak on Precision
Kimi K3's security benchmarks look competitive, but testing reveals a major precision gap and poor performance on large codebases. See the full results.

kimi-k3 is 1% behind Fable on DeepSWE XD

how does an open-weight model come that close to Fable on DeepSWE?

---

Regarding resolution: I haven't decided if it matches Fable/Mythos on its ability to find bugs, though I am leaning towards resolving "before 2027" to Yes

/notbayesian/an-open-weight-model-beats-opus-47

Free mana

Kimi K3 is already within 3 points on https://artificialanalysis.ai/

AI Model & API Providers Analysis | Artificial Analysis
Comparison and analysis of AI models and API hosting providers. Independent benchmarks across key performance metrics including quality, price, output speed & latency.

Which version of Fable/Mythos? Fixed version Mythos 5, Fable 5, Mythos Preview, or other?

@David6LScg release version of Fable/Mythos

People are also trading

Will OpenAI release a Mythos-class model before Fable is reenabled?
1% chance
Will Anthropic release an open-weights model in 2026?
10% chance
Weights in largest open-weight AI model before October 2026?
Weights in largest open-weight AI model before 2027?
When will Llama 4 Behemoth weights become publicly available?
10/14/26
Will a non-Anthropic company make a Mythos-class model generally available worldwide in 2026?
86% chance
An open weight model beats Opus 4.7 by EOY26
86% chance
Will "Mythos-class" model be called Mythos?
11% chance
Will GigaChat release an open-weights model with ≥100B parameters by the end of 2026?
45% chance
Will Anthropic require training-consent for any frontier model by mid-2027?
16% chance