An agent repeats context, calls tools and may produce discarded output before completing a task. Together’s current Muse Spark Open page, listing an October 15 release, prices input at $1.25 and output at $4.25 per million tokens, with cached input at $0.15.
Current impact
These rates make parts of the bill explicit, but they are not a price per successful task. The immediate opportunity is to measure how context reuse and output volume affect a real workflow. Caching benefits depend on eligible repeated input, not the total prompt size alone.
Potential greater impact
If a team can reuse context without sacrificing correctness, it could lower the cost of repeated agent work. The larger benefit depends on completion rates, retries, tool charges and latency. A cheaper session that requires more human repair may be more expensive overall.
What would establish progress
Compare representative tasks using accepted outcomes and total spend, recording the date and serving mode. Keep input, cache and output accounting separate. Published model rates are a starting point for operational budgeting, not a substitute for workload measurement.
Image: Together AI / official announcement. Used for source-attributed editorial reporting; rights remain with the owner.
Sources & evidence
Sources describe the announcements and evidence available when this analysis was published. Company claims are attributed; plans and trials are not treated as sustained operation.
