Example 1: one coding agent session, 150,000 input + 12,000 output tokens, standard rate, no cache Opus 4.1 input $2.25 + output $0.90 = $3.15 Opus 5 input $0.75 + output $0.30 = $1.05 Sonnet 5 input $0.30 + output $0.12 = $0.42 Fable 5.1 input $1.50 + output $0.60 = $2.10 Example 2: same session, 120,000 of the input tokens served as cache hits, 30,000 at base input Opus 4.1 cache hits $0.18 + fresh input $0.45 + output $0.90 = $1.53; saving vs example 1 $1.62 (51.4%); one-time 5m cache write of 120,000 tokens $2.25 Opus 5 cache hits $0.06 + fresh input $0.15 + output $0.30 = $0.51; saving vs example 1 $0.54 (51.4%); one-time 5m cache write of 120,000 tokens $0.75 Sonnet 5 cache hits $0.02 + fresh input $0.06 + output $0.12 = $0.20; saving vs example 1 $0.22 (51.4%); one-time 5m cache write of 120,000 tokens $0.30 Fable 5.1 cache hits $0.03 + fresh input $0.30 + output $0.60 = $0.93; saving vs example 1 $1.17 (55.7%); one-time 5m cache write of 120,000 tokens $1.50 Example 3: 200 such sessions in a month on the Batch API rate (50% off input and output), no cache Opus 4.1 batch input $225.00 + batch output $90.00 = $315.00; standard rate would be $630.00 Opus 5 batch input $75.00 + batch output $30.00 = $105.00; standard rate would be $210.00 Sonnet 5 batch input $30.00 + batch output $12.00 = $42.00; standard rate would be $84.00 Fable 5.1 batch input $150.00 + batch output $60.00 = $210.00; standard rate would be $420.00 Ratios Opus 4.1 / Opus 5 base input 3x, output 3x Opus 5 / Sonnet 5 base input 2.5x, output 2.5x, cache hit 2.5x Fable 5.1 / Opus 5 base input 2x, output 2x Fable 5.1 / Opus 4.1 base input 0.6667, output 0.6667 Fast mode Opus 5 / standard Opus 5 input 2x, output 2x Fable 5.1 cache hit / base input 0.025; Opus 5 cache hit / base input 0.1 Example 2 Fable 5.1 / Opus 5 cached session 1.82x (uncached example 1 ratio 2x)