Muse Code test burns about $230 in credits over 3 hours, user says
A user says four parallel Muse Spark 1.3 Max agents used up the 5-hour limit in 3 hours, calling the limits far more generous than Codex or Claude but noting the lack of a desktop app and an unusable voice mode.
I just burned through ~$230 worth of tokens on $META Muse Code It took me 3 hours to reach the 5h limit Here is my initial review: -> Model usage limits Very generous 5-hour limit! I got a lot of work done and it took 4 parallel agents running Muse Spark 1.3 Max for 3 hours to drain it 5 hour limit is roughly 1/3 of the weekly limit, which is nice to see (OpenAI, Anthropic take notes) I like this approach way better than Codex/Claude where just looking at Astra/Opus will drain your 5-hour limit -> The value proposition For the $15 plan you get: ~165M 5 hour token limit = $231 ~500M weekly tokens = $700 per week So roughly $2800 in API-equivalent usage for $15 lol Again, this is also better than Codex/Claude by miles (remember MS1.3 is 3-5x cheaper so these numbers carry more than it seems) The actual takeaway is you get to use frontier intelligence for more than 30 minutes -> Then for the actual model I did some UI work, some API integrations and migrations + some heavier logic tasks for historical stock performance tracking... And it just handled everything I threw at it Then I asked Claude Opus 5 and GPT 6 Astra to review it from 1-10 Claude + Opus: Code Quality: 8.5 Correctness: 8 Production readiness: 7 Codex + Astra: Code Quality: 7 Correctness: 6 Production readiness: 5 Both had valid comments in review, but nothing blocking/crucial tbf Overall amazing value. There is only one thing holding me back right now @alexandr_wang if you are reading this: 1. Please drop a desktop app for Muse Code. Running multiple agents in CLI with different terminals with no naming is messy. MVP Project/session sidebar + simple GUI for actions would help a lot 2. The voice mode is not working. Takes ages to type out prompts. Codex has nailed this best so far for native dictation imo Exited to see whats next