BenchLeader
OpenAIReasoning modelNewReleased 17 Nov 2026Compare with another model

GPT-6.2 Astra

Launch tracker · day 3

GPT-6.2 Astra at launch: the first independent results

As of 19 Nov 2026, GPT-6.2 Astra (OpenAI, released 17 Nov 2026) has independent results from Artificial Analysis (AA Intelligence Index v5.0.1 62.1), LMArena (Arena WebDev 1781), Epoch AI (GPQA Diamond 96.8%), Vals AI (Vals Index 67.7), LiveBench (LiveBench 2.0 84.9%), ARC Prize (ARC-AGI-2 95.8%), Scale AI (MultiChallenge 73.2%), Design Arena (Design Arena Website 1339) and Terminal-Bench (Terminal-Bench 62.1%). It ranks #2 of 343 models on the BenchLeader Index, at 68.9. It is listed at $8.00 in / $40.00 out per 1M tokens, with a 1.1M-token context window. Not yet measured by SWE-bench.

Released
17 Nov 2026 (day 3)
Provider
OpenAI
Price
$8.00 in / $40.00 out per 1M tokens
Context window
1.1M tokens
Open weights
No
BenchLeader Index
68.9 (#2 of 343 models)
Independent boardResults so far
Artificial AnalysisAA Intelligence Index v5.0.1 62.1 · MMMU-Pro 89.5% · AA-Omniscience v1.1 50 · more below
LMArenaArena WebDev 1781 · Arena Agent 15.2 · Arena Agent Task Completion 12.4
Epoch AIGPQA Diamond 96.8% · FrontierMath Tiers 1–3 92.6% · FrontierMath Tier 4 100.0% · more below
Vals AIVals Index 67.7 · BioMysteryBench 80.7% · Excel Modeling Benchmark v1.1 76.0% · more below
LiveBenchLiveBench 2.0 84.9% · LiveBench 2.0 Reasoning 91.6% · LiveBench 2.0 Coding 91.9% · more below
ARC PrizeARC-AGI-2 95.8% · ARC-AGI-1 98.5% · ARC-AGI-3 99.8%
Scale AIMultiChallenge 73.2% · PRBench Legal 43.1% · DrugDiscoveryBench 65.5% · more below
Design ArenaDesign Arena Website 1339 · Design Arena Agentic 1304
Terminal-BenchTerminal-Bench 62.1%

Boards usually post results within days of a launch; this page is rebuilt every three hours.

Scores from independent publishers only; benchmark figures a provider publishes about its own model are not shown.

Reasoning effort
Quality#2 of 351
68.9Index
Top 10%
+19.2 on the median
Price#249 of 262
$16.00per 1M tokens
Dearest quarter
18× the median
Speed#167 of 215
49tokens/s
Bottom quarter
0.6× the median
Value#249 of 262
4.3points per $
Bottom quarter
0.1× the median

Better value: Gemini 4 Neon scores 67.3 for $6.00/M, 63% cheaper.

No cheaper ranked model scores higher: it is on the price frontier.

As of 19 Nov 2026, GPT-6.2 Astra by OpenAI scores 68.9 on the BenchLeader Index (top 10 of 631 ranked configurations), with its best results in coding (76) and weakest in agents & tools (66). API pricing: $8.00 per 1M input tokens and $40.00 per 1M output tokens ($16.00 blended at 3:1), list price on 19 Nov 2026. It streams about 49 tokens/s (Artificial Analysis). Released 17 Nov 2026.

Key facts (as of 19 Nov 2026)

  • GPT-6.2 Astra API pricing: $8.00 per 1M input tokens and $40.00 per 1M output tokens ($16.00 blended at 3:1) (list price, 19 Nov 2026).
  • GPT-6.2 Astra streams about 49 tokens/s (Artificial Analysis, 19 Nov 2026).
  • GPT-6.2 Astra was released on 17 Nov 2026 (OpenAI).

In detail. GPT-6.2 Astra is an OpenAI proprietary reasoning model, released 17 Nov 2026. Its best configuration (max reasoning effort) ranks #2 of 631 on the BenchLeader Index at 68.9 ±2.3, in the top ten. The ± is the point: 19 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in coding (76) and lowest in agents & tools (66). At $16.00 per million tokens blended it is among the most expensive ranked models. Output speed of 49 tokens per second puts it in the slowest quarter, with a first answer in 202.7 s. It has been measured at 6 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 18 Nov 2026.

Blended price
$16.00/M
$8.00 in · $40.00 out
via Artificial Analysis
Output speed
49 tok/s
measured by Artificial Analysis
First answer
203 s
first token 3.58 s
Context
1.1M
via models.dev
Accepts
Text · Images · PDFs
via models.dev, OpenRouter
Cost per task
$4.00
14 min a task, AA Index
Tokens per task
41k
64% of it reasoning
Released
17 Nov 2026
How it scores by categoryDashed line = average model (50). One step of 15 = one standard deviation.
  1. Overall index69
  2. Reasoning73
  3. Coding76
  4. Agents & tools66
  5. Maths73
  6. Knowledge74
  7. Multimodal72
  8. Composite70

Embed this rank

More badge styles
[![GPT-6.2 Astra on BenchLeader: #2 overall](https://www.benchleader.com/badge/gpt-6-2-astra.svg?style=compact)](https://www.benchleader.com/models/gpt-6-2-astra)

Paste it at the top of README.md, beside your other badges. The badge redraws with every data update and links back to this page.

Get an email when OpenAI ships a new model

And once when it gets ranked, plus the weekly review on Mondays.

Choose providers and emails

Versions

OpenAI has shipped 2 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.

ModelReleasedIndexRank
NewGPT-6.2 Astramaxthis page17 Nov 202668.9#2
GPT-6 Astramax2 Sep 202666.9#9

Reasoning-effort configurations

The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.

EffortIndexRankSpeedFirst answerChat reply costCategories
low60.7#8146 tok/s2.84 s$0.015Agents & tools 59 · Coding 61 · Knowledge 65 · Multimodal 67 · Reasoning 68
medium65.6#1547 tok/s3.89 s$0.015Agents & tools 70 · Coding 75 · Knowledge 68 · Multimodal 68 · Reasoning 70
high65.6#1648 tok/s26 s$0.015Agents & tools 64 · Coding 76 · Knowledge 69 · Multimodal 70 · Reasoning 73
xhigh67.1#848 tok/s67 s$0.015Agents & tools 71 · Coding 78 · Knowledge 70 · Multimodal 70 · Reasoning 73
maxbest68.9#249 tok/s203 s$0.015Agents & tools 66 · Coding 76 · Composite 70 · Knowledge 74 · Maths 73 · Multimodal 72 · Reasoning 73
not stated––37 tok/s3.58 s$0.015Instruction following 67 · Knowledge 45

What more thinking costs

Turning the effort up buys index points and multiplies the bill. Here max costs 6.0× low for +8.2 on the index — 3.2 points per doubling of spend. Compare that with other models.

6468low: $0.662 per task, index 60.760.7low $0.662medium: $1.62 per task, index 65.665.6medium $1.62high: $2.07 per task, index 65.665.6high $2.07xhigh: $2.52 per task, index 67.167.1xhigh $2.52max: $4.00 per task, index 68.968.9max $4.00Index

Each point is one reasoning setting: its cost per Artificial Analysis index task (log scale) and its BenchLeader Index.

How its index has moved

17 Nov 2026 to 19 Nov 2026
626771
  • max
  • xhigh
  • medium
  • high

The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.

Benchmark results

See all LLM benchmarks. One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.

Reasoning

Benchmarklowmediumhighxhighmaxnot statedSourceTrend
GPQA Diamond––––96.8%#1–Epoch AI Benchmarking Hub
LiveBench 2.0 Reasoningnot in index––––91.6%#7–LiveBench
HLE-Diamond (AA)47.7%#4056.0%#2865.9%#1267.3%#1072.2%#4–Artificial Analysis
ARC-AGI-198.5%#198.5%#198.5%#198.5%#198.5%#1–ARC Prize
ARC-AGI-291.7%#1793.3%#894.6%#495.8%#195.8%#1–ARC Prize
ARC-AGI-395.1%#2497.8%#1199.8%#3100.0%#199.8%#3–ARC Prize
CritPt v2 hard subset (AA)61.0%#3667.5%#1871.5%#570.5%#971.5%#5–Artificial Analysis
MysteryMechanismnot in index––––53.1%#1–Vals AI
HLE-Diamond (Scale)not in indexeffort not stated–––––66.9%#2Scale AI SEAL
Chess Puzzlesnot in index––––78.0%#2–Epoch AI Benchmarking Hub
Mystery Game Puzzlesnot in index––––95.0%#1–Epoch AI Benchmarking Hub

Coding

Benchmarklowmediumhighxhighmaxnot statedSourceTrend
Arena WebDev––––1781#3–Arena (LMArena)
LiveBench 2.0 Codingnot in index––––91.9%#1–LiveBench
Vibe Code Bench v1.1not in index––––89.9%#9–Vals AI
Vibe Code Bench 1-100 v1.1not in index––––55.2%#1–Vals AI
SWE Atlas v1.1not in indexeffort not stated–––––60.5%#17Scale AI SEAL
SWE Atlas v1.1: Codebase QnAnot in indexeffort not stated–––––67.7%#15Scale AI SEAL
SWE Atlas v1.1: Test Writingnot in indexeffort not stated–––––62.2%#15Scale AI SEAL
SWE Atlas v1.1: Refactoringnot in indexeffort not stated–––––51.4%#19Scale AI SEAL
MirrorCodenot in index––––67.8%#10–Epoch AI Benchmarking Hub
Terminal-Bench 4.0 (AA)not in index51.0%#4358.6%#1960.1%#1164.1%#364.1%#3–Artificial Analysis
Design Arena Websitenot in indexeffort not stated–––––1339#5Design Arena
FrontierCode 2.0 (AA)32.7%#4446.6%#647.1%#549.0%#245.4%#8–Artificial Analysis

Agents & tools

BenchmarklowmediumhighxhighmaxSourceTrend
Terminal-Bench––––62.1%#18Terminal-Bench
Arena Agent––––15.2#6Arena (LMArena)
LiveBench Long Agenticnot in index––––70.8%#7LiveBench
LiveBench Browsernot in index––––93.0%#4LiveBench
Analyst Agent (AA)not in index56.3%#752.5%#1461.3%#362.5%#162.5%#1Artificial Analysis
Finance Agent v2.1not in index––––60.8%#26Vals AI
Harvey's Legal Agent Benchmark v1.1not in index––––16.3%#12Vals AI
Legal Research Benchnot in index––––42.8%#32Vals AI
Robo-benchnot in index––––32.7%#1Vals AI
Tax Agent Benchnot in index––––21.8%#54Vals AI
Terminal-Bench 4.0 (Vals)not in index––––63.1%#5Vals AI
Terminal-Bench Sciencenot in index––––71.4%#1Vals AI
Arena Agent Task Completionnot in index––––12.4#11Arena (LMArena)
SpireClimbnot in index––––41.3%#1Epoch AI Benchmarking Hub
AutomationBenchnot in index68.6%#3172.8%#1776.0%#1176.1%#1077.3%#7Artificial Analysis
GDP.pdf (AA)not in index28.0%#3428.4%#2928.6%#2528.0%#3429.8%#19Artificial Analysis
Harvey LAB-AA v1.1not in index2.2%#328.1%#1213.3%#312.8%#515.0%#2Harvey
EnterpriseOps-Gymnot in index21.8%#5922.6%#5825.2%#5525.3%#5429.9%#44Artificial Analysis
AA-Briefcase v1.1not in index1039#1651408#891570#371594#301643#17Artificial Analysis
Design Arena Agenticnot in index––––1304#9Design Arena
Terminal-Bench-Science 0.1 (AA)not in index51.0%#3174.8%#475.2%#378.6%#178.1%#2Artificial Analysis
CWE-Bench (AA)61.7%#1465.8%#665.8%#669.2%#265.0%#8Artificial Analysis
DeepSec-Bench (AA)39.4%#1056.4%#144.1%#952.8%#246.8%#6Artificial Analysis
CyberGym end-to-end (AA)not in index0.0%#220.0%#220.0%#220.0%#220.0%#22Artificial Analysis
MKA (AA)not in index––––52.7%#6Artificial Analysis

Knowledge

Benchmarklowmediumhighxhighmaxnot statedSourceTrend
SimpleQA Verified––––78.5%#2–Epoch AI Benchmarking Hub
LiveBench 2.0 Data Analysisnot in index––––84.1%#4–LiveBench
LiveBench Researchnot in index––––91.5%#5–LiveBench
AA-Omniscience v1.138.7#4044.8#1946.8#1448.3#1250#10–Artificial Analysis
PRBench Legaleffort not stated–––––43.1%#34Scale AI SEAL
BioMysteryBenchnot in index––––80.7%#3–Vals AI
Excel Modeling Benchmark v1.1not in index––––76.0%#5–Vals AI
DrugDiscoveryBenchnot in index–––65.5%#4––Scale AI SEAL
EBR-benchnot in index––––92.4%#1–Epoch AI Benchmarking Hub
AA-Omniscience v1.1: accuracynot in index60.6%#2864.2%#1566.3%#1067.2%#868.6%#6–Artificial Analysis
AA-Omniscience v1.1: non-hallucinationnot in index44.2%#9645.9%#9341.9%#10242.3%#9940.9%#104–Artificial Analysis

Instruction following

Benchmarknot statedSourceTrend
MultiChallenge73.2%#9Scale AI SEAL

Multimodal

BenchmarklowmediumhighxhighmaxSourceTrend
LiveBench Designnot in index––––82.8%#3LiveBench
MMMU-Pro85.1%#1786.1%#1187.7%#587.9%#489.5%#1Artificial Analysis
Humanity's Sixth Sensenot in index–––63.2%#4–Scale AI SEAL
Furniture Assemblynot in index––––89.2%#2Epoch AI Benchmarking Hub
VIALS-AAnot in index––––65.8%#2Handshake AI
AA-3Dnot in index––––1318#1Artificial Analysis

Long context

BenchmarklowmediumhighxhighmaxSourceTrend
MLCRnot in index30.6%#2832.8%#1835.6%#1137.8%#839.4%#4Artificial Analysis

Composite

Benchmarklowmediumhighxhighmaxnot statedSourceTrend
Epoch Capabilities Indexnot in indexeffort not stated–––––174.2#1Epoch AI Benchmarking Hub
LiveBench 2.0––––84.9%#3–LiveBench
AA Intelligence Index v5.0.1not in index45.6#4457#1460.1#761.6#562.1#3–Artificial Analysis
Vals Indexnot in index––––67.7#8–Vals AI
Vals Finance Indexnot in index––––57.3%#22–Vals AI
Vals RSI Indexnot in index––––29.0%#6–Vals AI
Vals Science Indexnot in index––––68.4%#1–Vals AI
Vals Legal Indexnot in index––––69.7%#22–Vals AI
AA Cyber Indexnot in index33.7#2640.8#1536.6#2240.6#1637.3#19–Artificial Analysis

Where it wins

Benchmarks where this configuration ranks in the top five of every configuration measured.

API pricing: where to run it

Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.

ProviderSpeedFirst tokenInput $/MOutput $/MContextQuantisation
Azure58 tok/s3.91 s$8.00$40.001.1M–
Azure51 tok/s2.88 s$8.80$44.001.1M–
Amazon Bedrock48 tok/s3.11 s$8.00$40.001.1M–
OpenAI46 tok/s3.58 s$16.00$80.001.1M–
Azure42 tok/s3.54 s$9.60$48.001.1M–
OpenAI36 tok/s5.21 s$8.00$40.001.1M–

Usage

How much GPT-6.2 Astra is used in practice, as the gateways that publish it report. Usage is not part of the Index. Most used models and labs.

Share of coding-agent turns, 30 days
0.01%
99% in Codex

Source: T3 Code via Slopalytics.

What a task costs

Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache at $0.400 per 1M. Reasoning tokens are not modelled.

WorkloadTokens in / outCostWith cachingTime
Chat reply400 / 300$0.015$0.0133.5 min
Summarize a 30-page report12,000 / 600$0.120$0.0523.6 min
Code edit6,000 / 1,500$0.108$0.0743.9 min
Agentic coding session60,000 / 4,000$0.640$0.2984.7 min
Structured extraction2,000 / 200$0.024$0.0133.4 min

More from OpenAI

New this month

Compare GPT-6.2 Astra with…

See also

Embed or cite

Show where GPT-6.2 Astra ranks on a model card or README. The badge redraws itself with every daily update, carries the data date and links back to this page.

[![GPT-6.2 Astra on BenchLeader: #2 overall](https://www.benchleader.com/badge/gpt-6-2-astra.svg)](https://www.benchleader.com/models/gpt-6-2-astra)

Cite this

The ranking changes daily, so a reference carries the data date. APA style:

BenchLeader. (2026, November 19). GPT-6.2 Astra on the BenchLeader Index (method v1.5) [Data set]. Retrieved November 19, 2026, from https://www.benchleader.com/models/gpt-6-2-astra

BibTeX:

@misc{benchleader_gpt_6_2_astra_20261119,
  author       = {{BenchLeader}},
  title        = {{GPT-6.2 Astra on the BenchLeader Index (method v1.5)}},
  year         = {2026},
  month        = nov,
  howpublished = {\url{https://www.benchleader.com/models/gpt-6-2-astra}},
  note         = {Data as of 2026-11-19}
}

Every badge style, the embeddable top-10 widget and how to cite the Index

Questions people ask

What are GPT-6.2 Astra's benchmark scores?
As of 19 Nov 2026, GPT-6.2 Astra (max effort) scores 68.9 ±2.3 on the BenchLeader Index, top 10, strongest in coding (76) and weakest in agents & tools (66). That rests on independent benchmark results only, and every score on this page links to its publisher.
How does GPT-6.2 Astra rank?
GPT-6.2 Astra (max effort) is #2 of 631 ranked configurations on 19 Nov 2026 (top 10). 19 other configurations score within its margin, so the exact position matters less than the band. The leader is Claude Fable 5.2 (max effort) at 71.5.
How much does GPT-6.2 Astra cost per 1M tokens?
As of 19 Nov 2026, GPT-6.2 Astra costs $8.00 per 1M input tokens and $40.00 per 1M output tokens, $16.00 blended (three input tokens per output token).
Is GPT-6.2 Astra better than Gemini 4 Carbon?
As of 19 Nov 2026, Gemini 4 Carbon (high) and GPT-6.2 Astra (max) are neck and neck on quality (68.0 vs 68.9 on the BenchLeader Index, inside the ±2.9 margin of error), and Gemini 4 Carbon (high) costs 2.0× less, which makes it the better buy for most workloads. The head-to-head page has the benchmark-by-benchmark detail.
When was GPT-6.2 Astra released?
OpenAI released GPT-6.2 Astra on 17 Nov 2026. BenchLeader has results for it at 6 reasoning-effort settings.

Data as of 19 Nov 2026. Compare these configurations.

Cite as: BenchLeader, “GPT-6.2 Astra: benchmarks, pricing, speed and rank”, data as of 19 Nov 2026, https://www.benchleader.com/models/gpt-6-2-astra.