Positive users praise GPT-5.6 variants for cost and token efficiency plus future scaling potential while negative users dismiss the ECI leaderboard claims as ridiculous due to closed weights, missing sources, and questionable metrics.
Based on 7 visible X reactions from 15 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@scaling01 @teortaxesTex I actually don't feel that GPT-5.6-Sol is more token efficient. Never reached the 5-Hour Limit that fast. That's crazy...
@teortaxesTex yeah Fable 5.1 will mog but GPT-5.6-Sol being more cost and token efficient than Fable is actually brutal
@scaling01 Closed weights, 2026 dates, no sources. This is getting ridiculous.
@teortaxesTex Imagine post train runs on 20T GPT 6
@scaling01 how’s the trend looking like
@teortaxesTex yeah Fable 5.1 will mog but GPT-5.6-Sol being more cost and token efficient than Fable is actually brutal
@teortaxesTex Imagine post train runs on 20T GPT 6
@scaling01 how’s the trend looking like
@teortaxesTex It is on the way.
Truly the greatest post-training machine on Earth but even OpenAI can't overcome the physics of LLMs Mythos is a stronger base. Fable has a higher floor. Who knows what ECI Fable 5.1 will get. OpenAI will need a bigger model to reach a higher ceiling https://twitter.com/scaling01/status/2076073566085259405
funny that we're back to the era of Opus 3 vs GPT-4T Opus was slower, more expensive, and frankly less useful but it had more SOVL, wit, verbal fluency, big model smell that's how it is with Sol vs Fable again paradigms behind both have evolved drastically, but not priorities
GPT-5.6-Sol has a higher ECI than Fable 5 and GPT-5.6-Terra beats Opus 4.8 https://x.com/scaling01/status/2076073566085259405/photo/1
Positive users praise GPT-5.6 variants for cost and token efficiency plus future scaling potential while negative users dismiss the ECI leaderboard claims as ridiculous due to closed weights, missing sources, and questionable metrics.
Based on 7 visible X reactions from 15 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@teortaxesTex It is on the way.
Truly the greatest post-training machine on Earth but even OpenAI can't overcome the physics of LLMs Mythos is a stronger base. Fable has a higher floor. Who knows what ECI Fable 5.1 will get. OpenAI will need a bigger model to reach a higher ceiling https://twitter.com/scaling01/status/2076073566085259405
funny that we're back to the era of Opus 3 vs GPT-4T Opus was slower, more expensive, and frankly less useful but it had more SOVL, wit, verbal fluency, big model smell that's how it is with Sol vs Fable again paradigms behind both have evolved drastically, but not priorities