MASH C THOMSON

[MASH C THOMSON @Mashcthomson]

@theo But astra review takes like 50 mins some fix codex harness @OpenAI

[MASH C THOMSON @Mashcthomson]

@theo Yeah. fable planning,opus building,astra reviewing

[MASH C THOMSON @Mashcthomson]

@theo And people were saying medium was better than high or xhigh while even the video by lydia @AnthropicAI sent in its news letter says its doesn't do well than high on medium. Well its more on task and context based conditions i guess

[MASH C THOMSON @Mashcthomson]

@theo @addyosmani @grok summarize video and gimme timestamps

[MASH C THOMSON @Mashcthomson]

@theo did u check what subagents it used?. and what effort levels they are?. I did too and did not mention what agents to use so it used sonnet 5 and haiku for most stuff. Its really good with orchestration

[MASH C THOMSON @Mashcthomson]

@naveen_saradhi_ @kimmonismus More like one week its anthropic and once every 6 months its openai

[MASH C THOMSON @Mashcthomson]

@kimmonismus But one major issue is that its exploration is capped . Opus 5 was bad especially due to its verbosity which filled its reasoning, context and inturn made the whole run worser but it goes a long way to explore and search otherways or other meanings etc. opus 5.5 doesn't.

[MASH C THOMSON @Mashcthomson]

@moon_yose @VictorTaelin I wanted to do this but sonnet 5 is still too verbose ans circling. I always keep one cheap subagent open always during long running tasks so i can ask it questions on what the main one is doing and explain etc. .The /btw command routes to same model so same token cost so i avoid

[MASH C THOMSON @Mashcthomson]

@VictorTaelin yeah, for subagents depending on task pairing opus5.5 with luna actually works way better.

[MASH C THOMSON @Mashcthomson]

@Leophieudieu @scottstts And here i am trying to make a point astra and fable should be on same level and not compared to just opus. I wonder what will happen when fable 6 comes out given anthropic is jumping numbers now as well.

[MASH C THOMSON @Mashcthomson]

@Xxi5olc @scottstts Nope class means everything for tokens. They don't simply put 2x something usage for nothing

[MASH C THOMSON @Mashcthomson]

@1st_knight @LexnLin html js lol

[MASH C THOMSON @Mashcthomson]

@LexnLin What era was this game from?. We need more deetS

[MASH C THOMSON @Mashcthomson]

@dangreenheck I hate first-person games.

[MASH C THOMSON @Mashcthomson]

@dangreenheck @cosier Make the car a transformable one that can fly and cruise the sea as well

[MASH C THOMSON @Mashcthomson]

@DannyLimanseta Yeah, claude might change the scope of gaming entirely. But can someone fix @Fortnite first its buggy as hell since the layoffs

[MASH C THOMSON @Mashcthomson]

@xikhar Dayum but still once GTA6 comes out its gonna change the game on how much human work is needed for experience regardless of agents. I hope the @expedition33 team is also working on something.

[MASH C THOMSON @Mashcthomson]

@wholyv nah its finally matching sol usage levels.

[MASH C THOMSON @Mashcthomson]

@designwkarthick dude get 20x its enough make it drive luna well if u got codex

[MASH C THOMSON @Mashcthomson]

@jun_song Idk Astra usage was way better than
fable usage for me. Sol can run for days but opus 5 didn't the 5.5 is kinda matching sol. only the intelligence is bad because of codex harness wasting on tools.Everything good about claude only comes down to claude code harness.