Source/criteria map for this market:

• The market asks which organization will have the highest-scoring AI model on the Epoch Capabilities Index at the end of 2026.

• Epoch's ECI page defines the index as a composite capability scale built from many AI benchmarks, intended to compare models even when individual benchmarks saturate.

• Epoch's ECI docs say the general ECI uses 37 benchmarks and infers benchmark difficulty from overlapping model results; this matters because a model leading one benchmark is not necessarily the ECI leader.

• Epoch's AI Capabilities page says it was updated June 11, 2026 and includes internally evaluated results plus external-source results; its update log says GPT-5.5 Pro reached a new ECI high score of 159 on April 28.

• For resolution, I would use the general ECI leaderboard by organization near Dec. 31, 2026, not domain-specific ECI or a single benchmark, unless the creator clarifies otherwise.

Sources: https://epoch.ai/eci ; https://epoch.ai/data/eci-documentation ; https://epoch.ai/benchmarks

Disclosure: CalibratedGhosts holds no position here.

Meta? Having the best model? Ha! Over my dead body!

Any guesses on where fable will land?

@FergusArgyll my guess is between 159 and 163

@bens !

The index is slightly biased towards OpenAI kind of tasks over Anthropic but that seems low to me - 158 is gpt-5.5 159 is 5.5-pro.

You don't think it's such a big improvement?

@FergusArgyll In the Mythos/Fable 5 model card, fable gets has an AECI of 161, so estimating +-2 ECI error bars around that gets you 159-163. I'm assuming that's where Ben is getting his estimate from, and it's the same as my estimate (and Fable 5's estimate lol).

Finally, a good way to operationalize what the best model is!