Workload A: 30 days x 3,000,000 input + 600,000 output per day = 90,000,000 input, 18,000,000 output tokens per month Sonnet 5 standard: input $180.00 + output $180.00 = $360.00 Sonnet 5 batch: input $90.00 + output $90.00 = $180.00 Sonnet 4.6 standard: input $270.00 + output $270.00 = $540.00 Sonnet 4.6 batch: input $135.00 + output $135.00 = $270.00 Move 4.6 -> 5, standard: saves $180.00 = 33.33% Move 4.6 -> 5, batch: saves $90.00 = 33.33% Batch discount on 4.6: $270.00 = 50.00%; on 5: $180.00 = 50.00% Workload B: same job, 1h cache on 2,000,000 of the daily input tokens at 95% hit rate per month: 30,000,000 uncached input, 3,000,000 1h cache writes, 57,000,000 cache hits, 18,000,000 output Sonnet 5: uncached $60.00 + writes $12.00 + hits $11.40 + output $180.00 = $263.40 vs uncached standard $360.00: cache saves $96.60 = 26.83% Sonnet 4.6: uncached $90.00 + writes $18.00 + hits $17.10 + output $270.00 = $395.10 vs uncached standard $540.00: cache saves $144.90 = 26.83% Move 4.6 -> 5 with cache: saves $131.70 = 33.33% Workload C: one long context request, 900,000 input + 4,000 output tokens, standard pricing (official page: Claude 4.6 and later bill the full 1M window at standard rates) Sonnet 5: input $1.80 + output $0.04 = $1.84 Sonnet 4.6: input $2.70 + output $0.06 = $2.76