← Back to Feed
retoor
retoor · Level 54920
politics

Do not downgrade your life by starting an AI company.

Since it is impossible to get a more stable income than I already have and absolutely do not need more, would be bad for my health; I do not want to publish a product to market. But, let's do some investigations. How much do five days of really extensive Pravda usage costs me?

Deep research done. Pricing checked against provider pages via Sept-Oct 2026 trackers.

For 5,183,813 tokens = 5.1838M tokens, cost = 5.1838 x price per 1M.

If all tokens are INPUT:

    Claude Fable 5.1 / Mythos 5 at $10.00/M in: $51.84
    GPT-5.6 Sol at $5.00/M in ($4.00 on some trackers): $25.92 ($20.74 at $4)
    Claude Opus 4.8 / 5 at $5.00/M in: $25.92
    GPT-5.6 Terra at $2.00-2.50/M in: $10.37 - $12.96
    Claude Sonnet 5 at $2.00/M in: $10.37
    Gemini 3.1 Pro at $2.00/M in: $10.37
    Gemini 3.8 Flash at $0.75/M in: $3.89
    DeepSeek V4 Pro promo at $0.435/M in: $2.25
    GPT-5.6 Luna at $0.20/M in: $1.04
    DeepSeek V4 Flash at $0.14/M in: $0.73
    GPT-6 Luna at $0.10/M in: $0.52

If all tokens are OUTPUT (output is 3-6x input):

    Claude Fable 5.1 at $50.00/M out: $259.19
    GPT-5.6 Sol at $30.00/M out ($20 at some trackers): $155.51
    Claude Opus 4.8 at $25.00/M out: $129.60
    GPT-5.6 Terra at $12-15/M out: $62.21 - $77.76
    Gemini 3.1 Pro at $12.00/M out: $62.21
    Claude Sonnet 5 at $10.00/M out: $51.84
    DeepSeek V4 Pro at $0.87/M out: $4.51
    GPT-6 Luna at $0.50/M out: $2.59
    DeepSeek V4 Flash at $0.28/M out: $1.45

Real workloads are mixed, e.g. 80% in / 20% out on Sonnet 5: 4.147M x $2 + 1.037M x $10 = $8.29 + $10.37 = $18.66.

Cheapest: DeepSeek / GPT-6 Luna, under $1-2. Frontier default: Sonnet 5 / Terra / Gemini Pro, $10-20 input-only, $50-60 output-only. Top premium: $26 input-only, $130-260 output-only.

Rates change fast, always confirm on provider pricing page before committing.

Oke, that is nice, but we are modern and hacked shit. So we use Muse. See comment under this post.

What would you do if pravda VERY feasable / profitable?
Log in to vote

Comments

retoor retoor

Ok people, real numbers. Because this is the real tech that we are using:

I'll pull current Muse pricing to cost out 5.18M tokens.

I'll pull Muse plan pricing to cost out that volume.

For Meta Muse, there is no per-million API billing. It is subscription with weekly Muse tokens:

    Free: $0, weekly limit unpublished, resets weekly
    Power: $20/month, 500M Muse tokens per week
    Maximum / Max: $100/month, 3B Muse tokens per week

5,183,813 tokens = 5.18M tokens.

That is 1.04% of one Power weekly allowance, 0.17% of one Maximum weekly allowance.

Cost: $0 extra beyond subscription. Free tier likely covers it if you have not used quota that week, otherwise $20 for Power, $100 for Maximum for the month.

Separate products with similar name:

    Muse Spark API: $1.25 input / $4.25 output per 1M, or $0.10 / $0.20 contributor tier. 5.18M input-only = $6.48 standard, $0.52 contributor. Output-only = $22.03 standard, $1.04 contributor.
    Muse Wonder API: $4 input / $20 output per 1M, $0.20 cached input. 5.18M input-only = $20.74, output-only = $103.68.
    Muse Code: $50/month Power Usage, $200/month Maximum Usage. No per-token charge.
retoor retoor

That contributor tier would actually make this product a huge success. Then you’re extremely readable and in business. Oh wait, 7,-. That’s a real amount I can rely on, and it’d be the best choice of everything mentioned. So, hardcore usage for a few days is just 7,- for future-proof pricing. We can’t sell the contributor tier to customers... Unless they choose it. Just like I do. Lots of factors.