GPT-5.6 cost vs performance
Nine benchmark families, 192 traceable points and references, and one important warning: token-price changes do not automatically equal task-cost changes.
Method.
Scores never change. Current mode applies Luna ×0.2 and Terra ×0.8 to recoverable launch coordinates, plus Sol ×0.8 PROJECTED from its input-rate reduction. Sol output fell 33.3%, and benchmark token/tool mixes are unpublished, so this is a conservative visual estimate—not an exact benchmark invoice.
Source scores unchanged
9 / 192
9 benchmarks · 192 points / references
Current API rates
Input / cached input / output · USD per 1M text tokens
Sol promotional rates: $4 / $0.40 / $20 · available at least through November 21, 2026.
9 benchmarks · 192 points / references
Source scores unchanged
Showing Aug 21 current estimate with a log x-axis. Scores are unchanged.
Agents’ Last Exam
Score (%) vs estimated API cost · Aug 21 current estimate
Terra and Luna use July 30 adjusted pricing. Because OpenAI says cost simulations may include tool-call charges, adjusted total cost is labeled estimated.
PROJECTED from current token pricing; exact task cost depends on token and tool mix
View accessible data
| Model | Effort | Estimated cost | Score | Source quality |
|---|---|---|---|---|
| GPT-5.6 Sol | P1 | $98.4 · PROJECTED | 37.5% | Digitized |
| GPT-5.6 Sol | P2 | $188 · PROJECTED | 44.8% | Digitized |
| GPT-5.6 Sol | P3 | $341 · PROJECTED | 49.3% | Digitized |
| GPT-5.6 Sol | P4 | $446 · PROJECTED | 52.1% | Digitized |
| GPT-5.6 Sol | P5 | $602 · PROJECTED | 53.6% | Digitized |
| GPT-5.6 Sol | Max | $855 · PROJECTED | 52.7% | Digitized |
| GPT-5.6 Terra | P1 | $64.0 | 40.3% | Digitized |
| GPT-5.6 Terra | P2 | $94.4 | 42.6% | Digitized |
| GPT-5.6 Terra | P3 | $202 | 46.3% | Digitized |
| GPT-5.6 Terra | P4 | $293 | 48.5% | Digitized |
| GPT-5.6 Terra | Max | $420 | 50.4% | Digitized |
| GPT-5.6 Luna | P1 | $2.4 | 30.8% | Digitized |
| GPT-5.6 Luna | P2 | $9.2 | 36.2% | Digitized |
| GPT-5.6 Luna | P3 | $25.8 | 45.4% | Digitized |
| GPT-5.6 Luna | P4 | $47.4 | 48.6% | Digitized |
| GPT-5.6 Luna | Max | $81.4 | 50.3% | Digitized |
| GPT-5.5 | P1 | $124 | 37.5% | Digitized |
| GPT-5.5 | P2 | $280 | 41.3% | Digitized |
| GPT-5.5 | P3 | $372 | 44.9% | Digitized |
| GPT-5.5 | XHigh | $585 | 46.9% | Digitized |
| Claude Opus 4.8 | P1 | $1,168 | 37.5% | Digitized |
| Claude Opus 4.8 | P2 | $1,700 | 38.8% | Digitized |
| Claude Opus 4.8 | P3 | $2,800 | 42.2% | Digitized |
| Claude Opus 4.8 | Max | $4,000 | 45.2% | Digitized |
| Claude Fable 5 | Max | $2,300 | 40.5% | Digitized |
| Gemini 3.1 Pro | Preview | $2,000 | 32.1% | Digitized |
Artificial Analysis Intelligence Index v4.1
Index score vs estimated API cost · Aug 21 current estimate
Launch chart coordinates are preserved; Terra ×0.8 and Luna ×0.2 in adjusted-price view.
PROJECTED from current token pricing; exact task cost depends on token and tool mix
View accessible data
| Model | Effort | Estimated cost | Score | Source quality |
|---|---|---|---|---|
| GPT-5.6 Sol | P1 | $194 · PROJECTED | 41.2 | Digitized |
| GPT-5.6 Sol | P2 | $286 · PROJECTED | 49.5 | Digitized |
| GPT-5.6 Sol | P3 | $480 · PROJECTED | 53.6 | Digitized |
| GPT-5.6 Sol | P4 | $777 · PROJECTED | 55.9 | Digitized |
| GPT-5.6 Sol | P5 | $1,246 · PROJECTED | 57.8 | Digitized |
| GPT-5.6 Sol | Max | $2,274 · PROJECTED | 58.9 | Digitized |
| GPT-5.6 Terra | P1 | $103 | 34.1 | Digitized |
| GPT-5.6 Terra | P2 | $126 | 40.5 | Digitized |
| GPT-5.6 Terra | P3 | $194 | 45.7 | Digitized |
| GPT-5.6 Terra | P4 | $389 | 49 | Digitized |
| GPT-5.6 Terra | P5 | $594 | 51.7 | Digitized |
| GPT-5.6 Terra | Max | $1,417 | 55 | Digitized |
| GPT-5.6 Luna | P1 | $8.58 | 26.7 | Digitized |
| GPT-5.6 Luna | P2 | $14.3 | 33.3 | Digitized |
| GPT-5.6 Luna | P3 | $20.0 | 38.1 | Digitized |
| GPT-5.6 Luna | Max | $174 | 51.2 | Digitized |
| GPT-5.5 | P1 | $186 | 35.5 | Digitized |
| GPT-5.5 | P2 | $357 | 43.6 | Digitized |
| GPT-5.5 | P3 | $871 | 50.5 | Digitized |
| GPT-5.5 | P4 | $1,657 | 53.3 | Digitized |
| GPT-5.5 | XHigh | $2,643 | 54.8 | Digitized |
| Claude Fable 5 | Max | $5,639 | 59.9 | Digitized |
| Claude Opus 4.8 | Max | $3,760 | 55.7 | Digitized |
| Gemini 3.5 Flash | Not specified | $1,038 | 50.2 | Digitized |
| Gemini 3.1 Pro | Preview | $811 | 46.5 | Digitized |
Artificial Analysis Coding Agent Index v1.1
Index score vs estimated API cost · Aug 21 current estimate
Includes the recovered reasoning-effort points for GPT-5.6 variants, GPT-5.5 and competitors.
PROJECTED from current token pricing; exact task cost depends on token and tool mix
View accessible data
| Model | Effort | Estimated cost | Score | Source quality |
|---|---|---|---|---|
| GPT-5.6 Sol | P1 | $396 · PROJECTED | 57.3 | Digitized |
| GPT-5.6 Sol | P2 | $499 · PROJECTED | 68.3 | Digitized |
| GPT-5.6 Sol | P3 | $857 · PROJECTED | 73.7 | Digitized |
| GPT-5.6 Sol | P4 | $1,182 · PROJECTED | 76.4 | Digitized |
| GPT-5.6 Sol | P5 | $1,492 · PROJECTED | 78.1 | Digitized |
| GPT-5.6 Sol | Max | $2,018 · PROJECTED | 80 | Digitized |
| GPT-5.6 Terra | P1 | $114 | 38.6 | Digitized |
| GPT-5.6 Terra | P2 | $146 | 52.5 | Digitized |
| GPT-5.6 Terra | P3 | $260 | 63.2 | Digitized |
| GPT-5.6 Terra | P4 | $456 | 71 | Digitized |
| GPT-5.6 Terra | P5 | $537 | 72.4 | Digitized |
| GPT-5.6 Terra | Max | $776 | 77.4 | Digitized |
| GPT-5.6 Luna | P1 | $27.1 | 35.6 | Digitized |
| GPT-5.6 Luna | P2 | $17.6 | 40.8 | Digitized |
| GPT-5.6 Luna | P3 | $35.3 | 57.6 | Digitized |
| GPT-5.6 Luna | P4 | $69.2 | 66.9 | Digitized |
| GPT-5.6 Luna | P5 | $90.8 | 69.8 | Digitized |
| GPT-5.6 Luna | Max | $111 | 74.6 | Digitized |
| GPT-5.5 | P1 | $346 | 44.1 | Digitized |
| GPT-5.5 | P2 | $420 | 56.6 | Digitized |
| GPT-5.5 | P3 | $963 | 69.7 | Digitized |
| GPT-5.5 | P4 | $970 | 71.7 | Digitized |
| GPT-5.5 | XHigh | $1,763 | 76.4 | Digitized |
| Claude Opus 4.8 | P1 | $1,080 | 67 | Digitized |
| Claude Opus 4.8 | Max | $2,534 | 72.5 | Digitized |
| Claude Fable 5 | Max | $3,864 | 77.2 | Digitized |
| Gemini 3.1 Pro | Preview | $664 | 42.7 | Digitized |
Terminal-Bench 2.1
Score (%) vs estimated API cost · Aug 21 current estimate
Cost curve is shown where a defensible launch x-coordinate was recoverable. Other official scores are shown as horizontal references rather than inventing costs.
PROJECTED from current token pricing; exact task cost depends on token and tool mix
View accessible data
| Model | Effort | Estimated cost | Score | Source quality |
|---|---|---|---|---|
| GPT-5.6 Sol | Low | $0.496 · PROJECTED | 76% | Digitized |
| GPT-5.6 Sol | Medium | $0.672 · PROJECTED | 80% | Digitized |
| GPT-5.6 Sol | High | $0.84 · PROJECTED | 84% | Digitized |
| GPT-5.6 Sol | XHigh | $1.02 · PROJECTED | 86.5% | Digitized |
| GPT-5.6 Sol | Max | $1.36 · PROJECTED | 88.8% | Digitized |
| GPT-5.6 Sol Ultra | Ultra | $4.16 · PROJECTED | 91.9% | Digitized |
| GPT-5.5 | Low | $0.35 | 73% | Digitized |
| GPT-5.5 | Medium | $0.55 | 78% | Digitized |
| GPT-5.5 | High | $1.15 | 83% | Digitized |
| GPT-5.5 | XHigh | $2.2 | 85.6% | Digitized |
| GPT-5.6 Terra | Not specified | Score only — no cost invented | 87.4% | Official reference |
| GPT-5.6 Luna | Not specified | Score only — no cost invented | 84.7% | Official reference |
| Claude Mythos 5 | Not specified | Score only — no cost invented | 88% | Official reference |
| Claude Fable 5 | Not specified | Score only — no cost invented | 83.1% | Official reference |
| Claude Opus 4.8 | Not specified | Score only — no cost invented | 78.9% | Official reference |
| Gemini 3.1 Pro | Not specified | Score only — no cost invented | 70.7% | Official reference |
DeepSWE v1.1
Score (%) vs estimated API cost · Aug 21 current estimate
This benchmark has the cleanest price update because the canonical leaderboard recorded the July 30 pricing change.
PROJECTED from current token pricing; exact task cost depends on token and tool mix
View accessible data
| Model | Effort | Estimated cost | Score | Source quality |
|---|---|---|---|---|
| GPT-5.6 Sol | Low | $0.856 · PROJECTED | 45.4% | Exact |
| GPT-5.6 Sol | Medium | $1.49 · PROJECTED | 61.1% | Exact |
| GPT-5.6 Sol | High | $2.78 · PROJECTED | 69.4% | Exact |
| GPT-5.6 Sol | XHigh | $3.76 · PROJECTED | 70.7% | Exact |
| GPT-5.6 Sol | Max | $6.71 · PROJECTED | 72.7% | Exact |
| GPT-5.6 Terra | Low | $0.344 | 24.1% | Exact |
| GPT-5.6 Terra | Medium | $0.464 | 35.1% | Exact |
| GPT-5.6 Terra | High | $0.904 | 53.8% | Exact |
| GPT-5.6 Terra | XHigh | $1.7 | 60.2% | Exact |
| GPT-5.6 Terra | Max | $3.96 | 69.6% | Exact |
| GPT-5.6 Luna | Low | $0.014 | 1.5% | Exact |
| GPT-5.6 Luna | Medium | $0.044 | 11.3% | Exact |
| GPT-5.6 Luna | High | $0.156 | 44.2% | Exact |
| GPT-5.6 Luna | XHigh | $0.308 | 56.9% | Exact |
| GPT-5.6 Luna | Max | $0.606 | 67.2% | Exact |
| GPT-5.5 | Low | $1.2 | 27% | Exact |
| GPT-5.5 | Medium | $2.75 | 54% | Exact |
| GPT-5.5 | High | $5.1 | 64.4% | Exact |
| GPT-5.5 | XHigh | $7.23 | 67% | Exact |
| Claude Fable 5 | Low | $3.76 | 59.6% | Exact |
| Claude Fable 5 | Medium | $6.09 | 65.4% | Exact |
| Claude Fable 5 | High | $9.18 | 68.6% | Exact |
| Claude Fable 5 | XHigh | $13.4 | 69.9% | Exact |
| Claude Fable 5 | Max | $21.6 | 69.7% | Exact |
| Claude Opus 4.8 | Low | $2.29 | 40.8% | Exact |
| Claude Opus 4.8 | Medium | $3.44 | 48.7% | Exact |
| Claude Opus 4.8 | High | $4.28 | 51.8% | Exact |
| Claude Opus 4.8 | XHigh | $8.01 | 54.4% | Exact |
| Claude Opus 4.8 | Max | $13.2 | 59% | Exact |
| Gemini 3.1 Pro | High | $2.14 | 11.8% | Exact |
BrowseComp
Score (%) vs estimated API cost · Aug 21 current estimate
BrowseComp uses tools heavily. Adjusted Terra/Luna points are explicitly marked approximate because tool charges are not broken out publicly.
PROJECTED from current token pricing; exact task cost depends on token and tool mix
View accessible data
| Model | Effort | Estimated cost | Score | Source quality |
|---|---|---|---|---|
| GPT-5.6 Sol | P1 | $0.2 · PROJECTED | 64.8% | Digitized |
| GPT-5.6 Sol | P2 | $0.4 · PROJECTED | 66% | Digitized |
| GPT-5.6 Sol | P3 | $1.68 · PROJECTED | 69.8% | Digitized |
| GPT-5.6 Sol | P4 | $2.4 · PROJECTED | 85% | Digitized |
| GPT-5.6 Sol | P5 | $3.52 · PROJECTED | 88% | Digitized |
| GPT-5.6 Sol | Max | $6.4 · PROJECTED | 90.4% | Digitized |
| GPT-5.6 Sol Ultra | Ultra | $9.76 · PROJECTED | 92.2% | Digitized |
| GPT-5.6 Terra | P1 | $0.2 | 65% | Approximate |
| GPT-5.6 Terra | P2 | $0.64 | 75% | Approximate |
| GPT-5.6 Terra | P3 | $1.28 | 82% | Approximate |
| GPT-5.6 Terra | P4 | $1.92 | 85% | Approximate |
| GPT-5.6 Terra | Max | $2.56 | 87.5% | Approximate |
| GPT-5.6 Luna | P1 | $0.020 | 45.5% | Approximate |
| GPT-5.6 Luna | P2 | $0.040 | 50% | Approximate |
| GPT-5.6 Luna | P3 | $0.13 | 65% | Approximate |
| GPT-5.6 Luna | P4 | $0.2 | 75% | Approximate |
| GPT-5.6 Luna | P5 | $0.36 | 80% | Approximate |
| GPT-5.6 Luna | Max | $0.48 | 83.3% | Approximate |
| GPT-5.5 | P1 | $1.4 | 35% | Digitized |
| GPT-5.5 | P2 | $2.5 | 58% | Digitized |
| GPT-5.5 | P3 | $19.0 | 78% | Digitized |
| GPT-5.5 | XHigh | $35.0 | 84.4% | Digitized |
| Claude Mythos 5 | Not specified | Score only — no cost invented | 88% | Official reference |
| Claude Mythos Preview | Not specified | Score only — no cost invented | 87.9% | Official reference |
| Claude Opus 4.8 | Not specified | Score only — no cost invented | 84.3% | Official reference |
| Gemini 3.1 Pro | Not specified | Score only — no cost invented | 85.9% | Official reference |
GDPval-AA v2
Elo vs estimated API cost · Aug 21 current estimate
OpenAI publishes the comparable scores, but individual cost coordinates are not recoverable in a trustworthy public table. The site shows every official score as a reference instead of inventing x-values.
View accessible data
| Model | Effort | Estimated cost | Score | Source quality |
|---|---|---|---|---|
| GPT-5.6 Sol | Not specified | Score only — no cost invented | 1,747.8 | Official reference |
| GPT-5.6 Terra | Not specified | Score only — no cost invented | 1,593 | Official reference |
| GPT-5.6 Luna | Not specified | Score only — no cost invented | 1,591.8 | Official reference |
| GPT-5.5 | Not specified | Score only — no cost invented | 1,493.7 | Official reference |
| Claude Fable 5 | Not specified | Score only — no cost invented | 1,759.6 | Official reference |
| Claude Opus 4.8 | Not specified | Score only — no cost invented | 1,600.1 | Official reference |
| Gemini 3.1 Pro | Not specified | Score only — no cost invented | 962.3 | Official reference |
| Gemini 3.5 Flash | Not specified | Score only — no cost invented | 1,348.8 | Official reference |
OSWorld 2.0
Score (%) vs estimated API cost · Aug 21 current estimate
Recovered model curves from the original cost chart. Terra ×0.8 and Luna ×0.2 in adjusted mode.
PROJECTED from current token pricing; exact task cost depends on token and tool mix
View accessible data
| Model | Effort | Estimated cost | Score | Source quality |
|---|---|---|---|---|
| GPT-5.6 Sol | P1 | $1.73 · PROJECTED | 10.1% | Digitized |
| GPT-5.6 Sol | P2 | $1.45 · PROJECTED | 20.9% | Digitized |
| GPT-5.6 Sol | P3 | $5.93 · PROJECTED | 42.3% | Digitized |
| GPT-5.6 Sol | P4 | $11.7 · PROJECTED | 49.8% | Digitized |
| GPT-5.6 Sol | P5 | $16.1 · PROJECTED | 56.1% | Digitized |
| GPT-5.6 Sol | Max | $21.2 · PROJECTED | 62.6% | Digitized |
| GPT-5.6 Terra | P1 | $0.328 | 5% | Digitized |
| GPT-5.6 Terra | P2 | $0.328 | 13.4% | Digitized |
| GPT-5.6 Terra | P3 | $1.35 | 22.5% | Digitized |
| GPT-5.6 Terra | P4 | $4.9 | 37% | Digitized |
| GPT-5.6 Terra | P5 | $8.35 | 46.3% | Digitized |
| GPT-5.6 Terra | Max | $11.8 | 50.2% | Digitized |
| GPT-5.6 Luna | P1 | $0.012 | 9.6% | Digitized |
| GPT-5.6 Luna | P2 | $0.104 | 18.3% | Digitized |
| GPT-5.6 Luna | P3 | $0.618 | 33.2% | Digitized |
| GPT-5.6 Luna | P4 | $1.2 | 42.1% | Digitized |
| GPT-5.6 Luna | Max | $1.55 | 45.6% | Digitized |
| GPT-5.5 | P1 | $0.29 | 3.1% | Digitized |
| GPT-5.5 | P2 | $1.58 | 16.4% | Digitized |
| GPT-5.5 | P3 | $8.22 | 37.2% | Digitized |
| GPT-5.5 | P4 | $11.4 | 40.7% | Digitized |
| GPT-5.5 | XHigh | $15.9 | 47.5% | Digitized |
| Claude Opus 4.8 | P1 | $12.9 | 47.1% | Digitized |
| Claude Opus 4.8 | P2 | $17.5 | 48.8% | Digitized |
| Claude Opus 4.8 | P3 | $20.8 | 49.2% | Digitized |
| Claude Opus 4.8 | P4 | $26.7 | 49.9% | Digitized |
| Claude Opus 4.8 | Max | $31.3 | 54.8% | Digitized |
AutomationBench
Score (%) vs estimated API cost · Aug 21 current estimate
GPT-5.6 Sol and Fable 5 cost/score pairs match the canonical Zapier leaderboard. Other OpenAI snapshot scores are shown without invented costs.
PROJECTED from current token pricing; exact task cost depends on token and tool mix
View accessible data
| Model | Effort | Estimated cost | Score | Source quality |
|---|---|---|---|---|
| GPT-5.6 Sol | Max | $0.8 · PROJECTED | 18.1% | Exact |
| Claude Fable 5 | Max | $2.03 | 17.4% | Exact |
| GPT-5.6 Terra | Not specified | Score only — no cost invented | 15.2% | Official reference |
| GPT-5.6 Luna | Not specified | Score only — no cost invented | 14.9% | Official reference |
| GPT-5.5 | Not specified | Score only — no cost invented | 12.9% | Official reference |
| Claude Opus 4.8 | Not specified | Score only — no cost invented | 15.5% | Official reference |
| Gemini 3.5 Flash | Not specified | Score only — no cost invented | 14.5% | Official reference |
Evidence map
Sources and limits
Pricing and announcements link to their official OpenAI pages. Individual benchmark points retain their exact, digitized, approximate, or score-only classification; the source package does not claim a per-point public URL.
Exact = tabulated or leaderboard value. Digitized = coordinate recovered from the source chart. Official reference = published score without an invented cost.
Greenbyte point of view
Use the lowest tier that clears the workflow target.
Luna for volume; Terra for headroom; Sol when frontier capability changes the outcome. That is a product judgment—not a universal prescription from OpenAI.
Discuss an AI product decision