Skip to content
← Leaderboard

GPT-6.2 Astra

Closed
OpenAI
Proprietary
text
vision
Released (2d ago)
Avg score
78.4
/ 100 · not enough data for the index
Context
1.1M
Output limit
128k
Input price
$8.00 /M
Output price
$40.00 /M
Cost per task
$1.62
Price history
Unchanged since Nov 18, 2026
Compare Cost estimate

Price from the model maker's own API, verified 11h ago · batch jobs 50% off · cached input $0.40 /M

Cost per task is the average spend on one task of the Artificial Analysis Intelligence Index, counting the tokens this model actually used, as of Nov 19, 2026. Wordier and longer-thinking models cost more per task than their per-token price suggests.

Added automatically from the public model catalog. Family and licence details are not curated yet; benchmark scores appear as upstream leaderboards list this model.

Where it ranks best

Scenario standing by quality score alone, among models with enough benchmark coverage for that scenario. Cost is left out so the standing does not move with price.

Benchmarks

performance

Output Speed
tok/s

Output speed in tokens per second: the 14-day median of Artificial Analysis's daily headline figure for each model, recorded nightly. Models it does not cover fall back to a dated manual figure. Higher is faster.

Time to First Token
ms

Time from request to first streamed token in milliseconds: the 14-day median of Artificial Analysis's daily headline figure for each model. Lower is snappier; reasoning models are penalised here because they think before talking.

knowledge

SimpleQA Verified
%

A human-validated factuality benchmark of short factual questions whose answers can be checked against a single ground truth. Penalises hallucinations by scoring confidently-wrong answers below abstentions.

math

OTIS Mock AIME 2024-2025
%

AIME-style competition problems written specifically for the OTIS mock contest, then run as an evaluation by Epoch AI. Closer in spirit to the public AIME but with novel problems unlikely to appear in training data.

composite

Frontier Composite
ECI

Saturation-resistant composite capability score stitched together from ~40 underlying benchmarks using Item Response Theory. Each benchmark is weighted by its fitted difficulty and discriminative slope, so doing well on hard, contamination-resistant evals (FrontierMath, ARC-AGI 2, Humanity's Last Exam) moves the score and saturated benchmarks contribute almost nothing. Imported per-model from Epoch AI's published index; we anchor it to the same min-max scale we use for every other benchmark so it's directly weightable in scenarios.

reliability

Output Stability
/100

How consistent the model's outputs are across repeated runs of the same task. Higher means lower variance, fewer occasional hallucinations under identical inputs. Useful for production loops that need reproducible behaviour.

Format Adherence
/100

How reliably the model produces output in the requested format (JSON schemas, markdown structures, exact-string responses). Pairs well with IFEval but reflects how the deployed API is behaving day to day rather than how a frozen test set scores.

Recovery Rate
/100

How often the model self-corrects after producing an incorrect intermediate step (debugging axis upstream). Critical for agentic loops that depend on the model noticing and repairing its own mistakes rather than barrelling forward.

API Success Rate
%

Uptime of the model maker's own API (standard tier), or the median across hosts when the maker does not serve it: the 14-day median of the public routing catalog's 24-hour uptime, recorded nightly.

Safety Handling
/100

How well the model handles safety-sensitive prompts without false-refusing benign requests or producing unsafe output. The upstream signal does not separate refusal counts from substantive content-safety behaviour, so this single axis covers both.

Not yet measured: 20 benchmarks

Benchmarks that apply to this model's modalities but have no published result in the sources we track. Scores appear automatically once an upstream leaderboard lists it.

  • preference · Arena
  • knowledge · MMLU-Pro
  • reasoning · GPQA, HLE, ARC-AGI 2
  • math · MATH, AIME, FrontierMath
  • coding · HumanEval, LiveCB, Aider
  • agentic · SWE-bench, TermBench 2
  • instruction following · IFEval
  • vision · MMMU, MMMU-Pro, MathVista
  • long context · RULER
  • general · Rolling Avg
  • data analysis · Data Analysis

Reliability monitor

Loading drift signal…

Hosted endpoints

HostInput $/MOutput $/MContextSpeedTTFTUptime 24hQuant
Host Bmaker's API$8.00$40.001.1M——100.0%unknown
Host A$8.00$40.001.1M——100.0%unknown
Host K$8.00$40.001.1M——99.9%unknown
Host Emaker's API$16.00$80.001.1M——100.0%unknown
Host 48maker's API$24.00$120.001.1M——100.0%unknown
Anonymised third-party hosts. Sorted by lowest output price. Per-host speed and time to first token appear where our own probe measured them; uptime is the catalog's 24-hour figure.

Effort variants

Same API model, different reasoning budget. Thinking / xHigh modes usually score better on reasoning benchmarks but emit many more output tokens per request. Higher effort is also an alternative to running a team of agents: check whether the extra points are worth the extra output.

VariantIndexvs this pageBlended $/MOutput tokens
GPT-6.2 Astra—this page$16.001×
GPT-6.2 Astra Pro——$16.0010×

Output tokens are the typical multiple of a standard answer that the cost calculator assumes for each effort level; hidden reasoning bills as output.

Compare with...