GPT-6 Astra pricing: real cost per million tokens
What OpenAI's GPT-6 Astra really costs across 5 plans from 3 providers: monthly fees, usable allowances and real per-token prices, against the official API list price and independent leaderboards.
Data updated · How real prices are computed
As of 2026-10-08, the cheapest way to use GPT-6 Astra is ChatGPT Pro 20x at a real cost of about $0.05247 per million tokens ($200/month for roughly 3.812B tokens a month). We track 5 plans and API channels that serve GPT-6 Astra.
- Lowest real price
- $0.05247/MTokChatGPT Pro 20x
- API list price
- —
- Most monthly tokens
- 3.812BChatGPT Pro 20x
- Plans & APIs
- 55 subscriptions · 0 metered APIs
Plot GPT-6 Astra on the price–capability chart →
GPT-6 Astra plans and real prices
Real price = monthly fee ÷ tokens you can actually use in a month. “vs API” shows how many times cheaper than the official API list price (standard workload mix) the plan is. “Evidence” opens the full calculation and sources for that row.
| Plan | Provider | Billing | Monthly fee | Monthly tokens | Real $/MTok | vs API | Confidence | Data date | |
|---|---|---|---|---|---|---|---|---|---|
| ChatGPT Pro 20x | OpenAI | Subscription | $200/mo | 3.812B | $0.05247 | — | Medium | Evidence ↓ | |
| ChatGPT Pro 5x | OpenAI | Subscription | $100/mo | 0.91B | $0.1099 | — | Low | Evidence ↓ | |
| ChatGPT Plus | OpenAI | Subscription | $20/mo | 0.159B | $0.1258 | — | High | Evidence ↓ | |
| Devin Max | Devin | Subscription | $200/mo | 1.115B | $0.1794 | — | Medium | Evidence ↓ | |
| Droid Max | Factory | Subscription | $200/mo | 1.01B | $0.1979 | — | Medium | Evidence ↓ |
Behind each number
For every plan: the data date, the exact calculation, the quota basis, how leaderboard scores were mapped, and the original adoption evidence with its sources.
ChatGPT Pro 20xOpenAI$0.0524659/MTok
- Plan
- ChatGPT Pro 20x · OpenAI
- Data date
- 2026-09-07 – 2026-09-21 · sampled
- Real price / MTok
- $0.0524659
- Monthly tokens
- 3,812,000,000 (3.812B)
- Quota confidence
- Medium
- Calculation
- $200 / month ÷ 3,812,000,000 tokens × 1,000,000 ≈ $0.0524659 / MTok
- Quota basis
Not workload-normalized: sample lacks a token breakdown (or is an official absolute token table); raw tokens used as-is
- Benchmark mapping
- Artificial Analysis Intelligence Index v4.3: Medium (5 configurations)
Terminal-Bench 4.0: Low (5 configurations)
Terminal-Bench 4.0 (AA): Low (5 configurations)
Code Arena | WebDev Overall: Low (1 configuration)
Agent Arena · Overall (Ranking / Pareto): Low (1 configuration)
Artificial Analysis Coding Agent Index v1.5: Low (2 configurations)
OpenDesign Arena · Quality ranking: Low (1 configuration)
DeepSWE v1.1: Low (5 configurations)Mapping confidence: how directly each score refers to the model served here.
Adoption evidence · source and rationale (translated from the Chinese record)
Weighted across 8 measured sources: Observatory 8.53, round10 screenshot 13.8, round13 same-frame 7.52, round14 maintainer panel 10.0, msg7086 8.07, round14 figure 4 (2/3 of a week) 8.18, figure 2 self-reported 9.25, figure 3 background 1.03B/week — the earlier figures share that scale; chatgpt-astra-round12/13/14
Added 3.812B: weekly pool 953M ×4 weeks — measured sources weighted by verification tier (same-frame panel / maintainer panel / continuous series ×3, self-reported share ×2, community verbal ×1); round10's 10% figure and tier, confirmed, changed from not-adopted to weighted-in; round14 new basis: weekly pool ≈$1200~1500 list-worth (Astra), same pool Sol $2200~2500, internal weighting penalizes Astra ~1.9×; the maintainer's own test ≈3.2B/month and lichengzhe's gateway 2.1B~2.3B are kept out of the weights per the ruling — pure verbal and lower-bound-only sources do not enter the mean; implied weight ≈3.38×Sol; same-kind real measurements still spread over 680M~1.56B/week — pools may genuinely differ across accounts, so this value is a weighted center, not a universal constant; not workload-normalized: eight-source weighting, most sources lack a token breakdown
ChatGPT Pro 5xOpenAI$0.10989/MTok
- Plan
- ChatGPT Pro 5x · OpenAI
- Data date
- 2026-07-30 – 2026-09-21 · derived; uses the data date of ChatGPT Plus · gpt-5.6-sol；ChatGPT Pro 20x · gpt-6-astra、gpt-5.6-sol
- Real price / MTok
- $0.10989
- Monthly tokens
- 910,000,000 (0.91B)
- Quota confidence
- Low
- Calculation
- $100 / month ÷ 910,000,000 tokens × 1,000,000 ≈ $0.10989 / MTok
- Quota basis
Not workload-normalized: sample lacks a token breakdown (or is an official absolute token table); raw tokens used as-is
- Benchmark mapping
- Artificial Analysis Intelligence Index v4.3: Medium (5 configurations)
Terminal-Bench 4.0: Low (5 configurations)
Terminal-Bench 4.0 (AA): Low (5 configurations)
Code Arena | WebDev Overall: Low (1 configuration)
Agent Arena · Overall (Ranking / Pareto): Low (1 configuration)
Artificial Analysis Coding Agent Index v1.5: Low (2 configurations)
OpenDesign Arena · Quality ranking: Low (1 configuration)
DeepSWE v1.1: Low (5 configurations)Mapping confidence: how directly each score refers to the model served here.
Adoption evidence · source and rationale (translated from the Chinese record)
Derived from same-plan gpt-5.6-sol 3.08B × 0.295458068516509
953M→910.011M: 38.12×30.8/129.02 (reusing the Sol 20x→5x inter-tier ratio; the baseline moves with the Sol 20x weighted value 4.19×); round12 codex#45085 prolite same-frame 231M/week ≈ 920M/month is close in magnitude but is a multi-agent Astra-High amplified sample — not directly adopted; chatgpt-astra-sameframe-round13-2026-09-20.json
chatgpt-astra-sameframe-round13-2026-09-20.jsonBrowse the public evidence archive
ChatGPT PlusOpenAI$0.125786/MTok
- Plan
- ChatGPT Plus · OpenAI
- Data date
- 2026-09-11 · sampled
- Real price / MTok
- $0.125786
- Monthly tokens
- 159,000,000 (0.159B)
- Quota confidence
- High
- Calculation
- $20 / month ÷ 159,000,000 tokens × 1,000,000 ≈ $0.125786 / MTok
- Quota basis
Not workload-normalized: sample lacks a token breakdown (or is an official absolute token table); raw tokens used as-is
- Benchmark mapping
- Artificial Analysis Intelligence Index v4.3: Medium (5 configurations)
Terminal-Bench 4.0: Low (5 configurations)
Terminal-Bench 4.0 (AA): Low (5 configurations)
Code Arena | WebDev Overall: Low (1 configuration)
Agent Arena · Overall (Ranking / Pareto): Low (1 configuration)
Artificial Analysis Coding Agent Index v1.5: Low (2 configurations)
OpenDesign Arena · Quality ranking: Low (1 configuration)
DeepSWE v1.1: Low (5 configurations)Mapping confidence: how directly each score refers to the model served here.
Adoption evidence · source and rationale (translated from the Chinese record)
Maintainer's Plus panel: 10,336,745 tokens (input+cache_read) = 26pp remaining of the weekly window; chatgpt-astra-adoption-round7-2026-09-11.json
Added 159M: 10,336,745÷26%×4 weeks; this extract excludes output (under Luna's method output is 0.33%, impact <1%); 26pp is a rounded-reading difference, range about 153M~165M; the Plus pricing page marks Astra as a limited tier (credits can be added) — what was directly measured is the actual consumption rate, which is unaffected; round8 found Observatory now measures Astra≈4.1×Sol, so round7's old 2× weight corroboration is doubtful and this value does not rely on the weight model; round13 added Plus same-frame samples of 32~75M/week spread over 128M~300M/month, same magnitude as 159M — value unchanged; Pro 20x already anchored per round13; not workload-normalized: the extract contains only input+cache_read, lacking the output component
chatgpt-astra-adoption-round7-2026-09-11.jsonBrowse the public evidence archive
Devin MaxDevin$0.179372/MTok
- Data date
- 2026-09-14 · sampled
- Real price / MTok
- $0.179372
- Monthly tokens
- 1,115,000,000 (1.115B)
- Quota confidence
- Medium
- Calculation
- $200 / month ÷ 1,115,000,000 tokens × 1,000,000 ≈ $0.179372 / MTok
- Quota basis
Standard workload: 97% cache reads / 2.5% input / 0.5% output
- Benchmark mapping
- Artificial Analysis Intelligence Index v4.3: Medium (5 configurations)
Terminal-Bench 4.0: Low (5 configurations)
Terminal-Bench 4.0 (AA): Low (5 configurations)
Code Arena | WebDev Overall: Low (1 configuration)
Agent Arena · Overall (Ranking / Pareto): Low (1 configuration)
Artificial Analysis Coding Agent Index v1.5: Low (2 configurations)
OpenDesign Arena · Quality ranking: Low (1 configuration)
DeepSWE v1.1: Low (5 configurations)Mapping confidence: how directly each score refers to the model served here.
Adoption evidence · source and rationale (translated from the Chinese record)
Maintainer's Devin Max panel cc usage: this week gpt-6-astra-high total 305,025,580 tokens (calls 667, in 1,998 / out 369,918 / cache_read 300,944,710 / cache_create 3,708,954) = 87pp of the weekly quota (remaining 100%→13%); devin-usage-round4-2026-09-14.json; anthropic-token-mix-round1-2026-09-24.json; https://devin.ai/pricing Max $200/month
1.402B→1.115B (2026-09-24 decision: normalized to the unified workload): segment worth $356.55 (cache read 300.945M×$1 + write/input 3.711M×$10 + output 0.370M×$50; OpenAI has no cache-write fee, so cache_create is billed as regular input) ÷87%×4 weeks = $1,639.31/month list-worth ÷ standard-workload blended price $1.47/MTok; the panel-% rounding interval is about 1.103B~1.128B; the raw-total basis 305,025,580÷87%×4 weeks = 1.402B is kept for comparison; 87pp is a near-full-week sample (3.76× round2's 20pp segment), superseding the old back-derivation — raw weekly pool 406M→350.6M (−13.7%, round2/3 kept as historical evidence); hit rate 98.78% under the cache_create-inclusive basis (same magnitude as round2's 97.84% — an extreme cache-heavy workload); swe-2-max and other free models do not draw down quota; the normalization assumes Devin deducts quota in list-price proportion; the Pro $20 tier has no data and is not derived
https://devin.ai/pricinganthropic-token-mix-round1-2026-09-24.jsondevin-usage-round4-2026-09-14.jsonBrowse the public evidence archive
Droid MaxFactory$0.197941/MTok
- Data date
- 2026-09-29 · derived; uses the data date of Droid Max · claude-opus-5.5
- Real price / MTok
- $0.197941
- Monthly tokens
- 1,010,400,000 (1.01B)
- Quota confidence
- Medium
- Calculation
- $200 / month ÷ 1,010,400,000 tokens × 1,000,000 ≈ $0.197941 / MTok
- Quota basis
Anthropic workload: 97% cache reads / 2.5% cache writes / 0.5% output
- Benchmark mapping
- Artificial Analysis Intelligence Index v4.3: Medium (5 configurations)
Terminal-Bench 4.0: Low (5 configurations)
Terminal-Bench 4.0 (AA): Low (5 configurations)
Code Arena | WebDev Overall: Low (1 configuration)
Agent Arena · Overall (Ranking / Pareto): Low (1 configuration)
Artificial Analysis Coding Agent Index v1.5: Low (2 configurations)
OpenDesign Arena · Quality ranking: Low (1 configuration)
DeepSWE v1.1: Low (5 configurations)Mapping confidence: how directly each score refers to the model served here.
Adoption evidence · source and rationale (translated from the Chinese record)
Derived from same-plan claude-opus-5.5 2.526B × 0.4
Factory official multipliers Opus 5.5 1.6× / gpt-6-astra 4×: the same Standard Usage pool scaled by multiplier — not a measurement of this model; inherits the Opus 5.5 row's Anthropic-tier workload, not recomputed for each model's own price gap; docs.factory.ai/docs/models; droid-model-multipliers-2026-09-29.json
droid-model-multipliers-2026-09-29.jsonBrowse the public evidence archive
GPT-6 Astra benchmark scores
Highest archived configuration per leaderboard; ranks are among all models each board archives in this dataset. Scores are never mixed across boards.
| Leaderboard | Metric | Score | Rank | Configuration | Effort | Harness | Configs | Snapshot |
|---|---|---|---|---|---|---|---|---|
| Artificial Analysis Intelligence Index v4.3 | Intelligence Index | 52.67 | #4 / 60 | GPT-6 Astra (max) | Max | — | 5 | 2026-10-01 |
| Terminal-Bench 4.0 | Resolution Rate % | 58.18 | #2 / 18 | Codex - GPT-6 Astra (max) | Max | Codex | 5 | 2026-09-21 |
| Terminal-Bench 4.0 (AA) | Resolution Rate % | 59.6 | #2 / 53 | Artificial Analysis - GPT-6 Astra (xhigh) | xhigh | Artificial Analysis | 5 | 2026-10-01 |
| Code Arena | WebDev Overall | Arena Score | 1,789 | #2 / 54 | gpt-6-astra-max | Max | — | 1 | 2026-09-30 |
| Agent Arena · Overall (Ranking / Pareto) | Net Improvement % | 12.18 | #3 / 39 | GPT 6 Astra (Max) | Max | — | 1 | 2026-09-30 |
| Artificial Analysis Coding Agent Index v1.5 | Coding Agent Index | 61.65 | #5 / 19 | Codex - GPT-6 Astra (max) | Max | Codex | 2 | 2026-10-01 |
| OpenDesign Arena · Quality ranking | Average task score | 82.7 | #7 / 16 | GPT-6 Astra | — | OpenDesign | 1 | 2026-10-01 |
| DeepSWE v1.1 | Pass@1 % | 74.12 | #2 / 33 | mini-swe-agent - gpt-6-astra (xhigh) | xhigh | mini-swe-agent | 5 | 2026-09-03 |
Every archived configuration
Artificial Analysis Intelligence Index v4.3 5 configurations
| Configuration | Score | Score interval | Harness / effort / mode | Source task cost (mean / median) | Checked | Links |
|---|---|---|---|---|---|---|
| GPT-6 Astra (max) | 52.6737 | — | — / Max / — | — / — | 2026-10-01 | Source · Archive |
| GPT-6 Astra (xhigh) | 52.3863 | — | — / xhigh / — | — / — | 2026-10-01 | Source · Archive |
| GPT-6 Astra (high) | 50.9191 | — | — / High / — | — / — | 2026-10-01 | Source · Archive |
| GPT-6 Astra (medium) | 49.5704 | — | — / Medium / — | — / — | 2026-10-01 | Source · Archive |
| GPT-6 Astra (low) | 45.7819 | — | — / Low / — | — / — | 2026-10-01 | Source · Archive |
Terminal-Bench 4.0 5 configurations
| Configuration | Score | Score interval | Harness / effort / mode | Source task cost (mean / median) | Checked | Links |
|---|---|---|---|---|---|---|
| Codex - GPT-6 Astra (max)Official Terminal-Bench 4.0 leaderboard row; score belongs to this harness × model × effort configuration. | 58.18 | 55.39 – 60.97 | Codex / Max / — | — / — | 2026-10-01 | Source · Archive |
| Codex - GPT-6 Astra (xhigh)Official Terminal-Bench 4.0 leaderboard row; score belongs to this harness × model × effort configuration. | 57.88 | 55.16 – 60.6 | Codex / xhigh / — | — / — | 2026-10-01 | Source · Archive |
| Codex - GPT-6 Astra (high)Official Terminal-Bench 4.0 leaderboard row; score belongs to this harness × model × effort configuration. | 57.88 | 54.91 – 60.85 | Codex / High / — | — / — | 2026-10-01 | Source · Archive |
| Codex - GPT-6 Astra (medium)Official Terminal-Bench 4.0 leaderboard row; score belongs to this harness × model × effort configuration. | 54.24 | 51.58 – 56.9 | Codex / Medium / — | — / — | 2026-10-01 | Source · Archive |
| Codex - GPT-6 Astra (low)Official Terminal-Bench 4.0 leaderboard row; score belongs to this harness × model × effort configuration. | 50.61 | 47.86 – 53.36 | Codex / Low / — | — / — | 2026-10-01 | Source · Archive |
Terminal-Bench 4.0 (AA) 5 configurations
| Configuration | Score | Score interval | Harness / effort / mode | Source task cost (mean / median) | Checked | Links |
|---|---|---|---|---|---|---|
| Artificial Analysis - GPT-6 Astra (xhigh)AA's own-harness independent measurement of Terminal-Bench 4.0 (the terminalBench40 field in the models-board payload, fraction×100); scored separately from the official tbench.ai board. | 59.596 | — | Artificial Analysis / xhigh / — | — / — | 2026-10-01 | Source · Archive |
| Artificial Analysis - GPT-6 Astra (max)AA's own-harness independent measurement of Terminal-Bench 4.0 (the terminalBench40 field in the models-board payload, fraction×100); scored separately from the official tbench.ai board. | 59.0909 | — | Artificial Analysis / Max / — | — / — | 2026-10-01 | Source · Archive |
| Artificial Analysis - GPT-6 Astra (high)AA's own-harness independent measurement of Terminal-Bench 4.0 (the terminalBench40 field in the models-board payload, fraction×100); scored separately from the official tbench.ai board. | 54.0404 | — | Artificial Analysis / High / — | — / — | 2026-10-01 | Source · Archive |
| Artificial Analysis - GPT-6 Astra (medium)AA's own-harness independent measurement of Terminal-Bench 4.0 (the terminalBench40 field in the models-board payload, fraction×100); scored separately from the official tbench.ai board. | 49.4949 | — | Artificial Analysis / Medium / — | — / — | 2026-10-01 | Source · Archive |
| Artificial Analysis - GPT-6 Astra (low)AA's own-harness independent measurement of Terminal-Bench 4.0 (the terminalBench40 field in the models-board payload, fraction×100); scored separately from the official tbench.ai board. | 41.9192 | — | Artificial Analysis / Low / — | — / — | 2026-10-01 | Source · Archive |
Code Arena | WebDev Overall 1 configuration
Agent Arena · Overall (Ranking / Pareto) 1 configuration
Artificial Analysis Coding Agent Index v1.5 2 configurations
| Configuration | Score | Score interval | Harness / effort / mode | Source task cost (mean / median) | Checked | Links |
|---|---|---|---|---|---|---|
| Codex - GPT-6 Astra (max)Score belongs to this named agent/harness × model configuration; it is not a harness-independent model measurement. indexScore stored as percent (raw fraction × 100). | 61.645 | — | Codex / Max / — | $7.467 / — | 2026-10-01 | Source · Archive |
| Devin Fusion CLI - GPT-6 Astra XHigh + SWE-2 MediumScore belongs to this named agent/harness × model configuration; it is not a harness-independent model measurement. indexScore stored as percent (raw fraction × 100). | 58.8884 | — | Devin Fusion CLI / xhigh / — | $4.537 / — | 2026-10-01 | Source · Archive |
OpenDesign Arena · Quality ranking 1 configuration
DeepSWE v1.1 5 configurations
| Configuration | Score | Score interval | Harness / effort / mode | Source task cost (mean / median) | Checked | Links |
|---|---|---|---|---|---|---|
| mini-swe-agent - gpt-6-astra (xhigh) | 74.115 | 71.25 – 76.98 | mini-swe-agent / xhigh / — | $6.524 / $5.672 | 2026-09-12 | Source · Archive |
| mini-swe-agent - gpt-6-astra (high) | 73.2301 | 69.807 – 76.654 | mini-swe-agent / High / — | $5.724 / $4.962 | 2026-09-12 | Source · Archive |
| mini-swe-agent - gpt-6-astra (max) | 73.2301 | 72.4 – 74.06 | mini-swe-agent / Max / — | $12.37 / $11.04 | 2026-09-12 | Source · Archive |
| mini-swe-agent - gpt-6-astra (medium) | 72.7876 | 70.198 – 75.377 | mini-swe-agent / Medium / — | $4.38 / $3.713 | 2026-09-12 | Source · Archive |
| mini-swe-agent - gpt-6-astra (low) | 67.0354 | 65.735 – 68.336 | mini-swe-agent / Low / — | $2.189 / $1.721 | 2026-09-12 | Source · Archive |
These are reference mappings, not benchmarks of a subscription or API channel. Harness and quota-measurement effort alignment are unverified. Source task costs are not subscription task costs.
- Exact served-model reference; quota-measurement effort is unverified.
- Exact served-model reference only; product harness and quota-measurement effort are unverified. Not a benchmark measurement of this subscription or API channel.
Frequently asked questions
How much does GPT-6 Astra really cost per million tokens?
Measured as monthly fee ÷ tokens you can actually use, GPT-6 Astra ranges from $0.05247 (ChatGPT Pro 20x) to $0.1979 (Droid Max) per million tokens.
Which subscription gives the most GPT-6 Astra usage?
ChatGPT Pro 20x allows roughly 3.812B tokens of GPT-6 Astra per month for $200.
How does GPT-6 Astra score on independent leaderboards?
GPT-6 Astra has scores on 8 leaderboards; its strongest placement is 1,789 (Arena Score) on Code Arena | WebDev Overall, #2 of 54 models tracked.
Who makes GPT-6 Astra, and where can I use it?
GPT-6 Astra (model ID gpt-6-astra) is made by OpenAI and is available through Devin, Factory, OpenAI.
More OpenAI models
| Model | Lowest real $/MTok | Cheapest plan | Plans |
|---|---|---|---|
| GPT 5.6 Luna | $0.00142 | ChatGPT Pro 20x | 7 |
| GPT-6 Luna | $0.00198 | Droid Max | 4 |
| GPT 5.6 Terra | $0.00801 | ChatGPT Pro 20x | 5 |
| GPT 5.6 Sol | $0.0155 | ChatGPT Pro 20x | 6 |
| GPT-6 Sol | $0.01818 | ChatGPT Plus | 2 |
| GPT 5.5 | $0.02004 | ChatGPT Pro 20x | 3 |
| GPT-6.1 Sol | $0.02299 | ChatGPT Plus | 1 |