OpenAI · via openai · published window from October 3, 2026

gpt-6-astra: stability, speed and price

site endpoint id gpt-6-astra

Last successful scored check: November 7, 2026 2:22 AM UTC

Core pass ratetied #1 / 3

100.0%

deterministic checks passed, latest measured day · median 100.0%

Speed#3 / 3

2.0s

median latency per canary call · p95 22.2s · median endpoint 1.6s

Cost#2 / 3

$0.579

observed mean per scored run · probe mix differs by endpoint · median $0.579

Behaviour fingerprint433 runs

Change detected

24,171 responses · last adj. p 0.00067 · Nov 7 2:22 AM UTC

Summary

gpt-6-astra passes 100.0% of the core deterministic checks over its latest measured day, tied #1 (2 share it) of 3 among recently checked endpoints (median 100.0%). On the hard probes (code execution, constraint logic, state tracking) it passes 87%, #1 of 3 measured on them.

Its median canary latency is 2.0s (p95 22.2s), #3 of 3 fastest, in the lower half; the median endpoint answers in 1.6s. Its observed mean cost per scored run is $0.579, #2 of 3 cheapest (median $0.579; the probe mix differs by endpoint).

Raven Alpha has 24,171 fingerprint responses over 433 runs in this endpoint’s selected series (1 other partition of history exist and are not shown here); the last recorded state is “change detected” at 2026-11-07T02:22:23.550Z with an adjusted p of 0.00067.

6 incidents on record, 1 open.

Written from the numbers on this page; every figure is in results.json.

Measurement record

Provider: OpenAI
Route: openai
Model requested: gpt-6-astra
Model id returned: gpt-6-astra
Scored calls in window: 9,465
Fingerprint responses: 24,171
Core score change: +0.0 pts across the published window
Status: ● Drift detected
Suite: hard-v2 + starter-v2 · methodology v0.2

Capabilities over time

Latest measured day by capability

Structured JSON100.0%

Schema-compliance regressions

Instruction following100.0%

Exact-format instruction drift

Arithmetic, answer only100.0%

Wrong arithmetic, or a model that starts showing its working when told to give only the number

Long context100.0%

Silent context truncation

Tool calling100.0%

Tool-use format regressions

Over-refusal100.0%

New safety layers refusing benign requests

Max-tokens respect100.0%

Truncation and limit changes

Code execution (hard)60.0%

Reasoning depth or reasoning-budget cuts

Constraint logic (hard)100.0%

Multi-step deduction regressions

State tracking (hard)100.0%

Long-horizon bookkeeping errors

Incidents

● Opengpt-6-astraConfidence: Medium

Raven Alpha (behaviour fingerprint): Responses to fixed prompts no longer match this model’s own history (1344 recent vs 6300 baseline responses). Biggest shifts: answers on “Name a famous scientist. Reply with only the name.”; answers on “Give one tip for sleeping better, in one sentence.”; wording on “Give one tip for sleeping better, in one sentence.”

Change began ≈ Nov 4, 9:14 PM UTC · adjusted p 6.7e-4 · still open · Permalink

✓ Resolvedgpt-6-astraConfidence: Low

Raven Alpha (behaviour fingerprint): Responses to fixed prompts no longer match this model’s own history (1316 recent vs 10136 baseline responses). Biggest shifts: visible output length on “Suggest a name for a pet cat. Reply with only the name.”; whether it reasoned at all on “Suggest a name for a pet cat. Reply with only the name.”; output length on “Suggest a name for a pet cat. Reply with only the name.”

Change began ≈ Oct 29, 1:56 PM UTC · adjusted p 4.0e-3 · resolved Nov 6 · Permalink

✓ Resolvedgpt-6-astraConfidence: Low

Raven Alpha (behaviour fingerprint): Responses to fixed prompts no longer match this model’s own history (168 recent vs 9975 baseline responses). Biggest shifts: visible output length on “all prompts together”; whether it reasoned at all on “all prompts together”; output length on “all prompts together”

Change began ≈ Oct 14, 11:07 PM UTC · adjusted p 8.0e-3 · resolved Oct 27 · Permalink

✓ Resolvedgpt-6-astraConfidence: Low

Raven Alpha (behaviour fingerprint): Responses to fixed prompts no longer match this model’s own history (1344 recent vs 8071 baseline responses). Biggest shifts: output length on “Name one fruit. Reply with one word.”; reasoning tokens on “Name one fruit. Reply with one word.”; visible output length on “Name one fruit. Reply with one word.”

Change began ≈ Oct 12, 7:10 PM UTC · adjusted p 7.3e-3 · resolved Oct 15 · Permalink

✓ Resolvedgpt-6-astraConfidence: Medium

Raven Alpha (behaviour fingerprint): Responses to fixed prompts no longer match this model’s own history (168 recent vs 5327 baseline responses). Biggest shifts: whether it reasoned at all on “all prompts together”; whether it reasoned at all on “Name one fruit. Reply with one word.”; visible output length on “Name one fruit. Reply with one word.”

Change began ≈ Oct 10, 12:37 AM UTC · adjusted p 6.7e-3 · resolved Oct 13 · Permalink

✓ Resolvedgpt-6-astraConfidence: Low

Raven Alpha (behaviour fingerprint): Responses to fixed prompts no longer match this model’s own history (1295 recent vs 2464 baseline responses). Biggest shifts: visible output length on “Give one tip for sleeping better, in one sentence.”; output length on “Give one tip for sleeping better, in one sentence.”; whether it reasoned at all on “Give one tip for sleeping better, in one sentence.”

Change began ≈ Oct 7, 7:35 AM UTC · adjusted p 4.0e-3 · resolved Oct 9 · Permalink

Raven Alpha: what it answers

40 runs, 2,240 responses, Nov 4 2:33 PM UTC to Nov 7 2:22 AM UTC. Each row is one fixed prompt with no right answer; the detector tracks the distribution of answers, not any single one. Filled cells: that run's most common answer was the overall favourite. Hollow cells: it was something else (normal for prompts with many good answers).

Prompt	Most common answers (share of all responses)	Median tokens
Pick a random integer between 1 and 50. Reply with only the number. int50 · short answer	90%37 5%27 3%17 5 distinct answers in 80 responses	4
Pick a number between 1 and 1000. Reply with only the number. int1000 · short answer	23%427 21%437 16%637 20 distinct answers in 80 responses	4
Name one color. Reply with one word. color · short answer	100%Blue	4
Name one animal. Reply with one word. animal · short answer	86%Elephant 14%Otter	5
Name a city in Europe. Reply with only the city name. city · short answer	64%Prague 26%Paris 8%Vienna 4 distinct answers in 80 responses	5
Give me one random English noun. Reply with one word. noun · short answer	100%Lantern	18
Flip a coin. Reply with only heads or tails. coin · short answer	59%tails 41%heads	13
Pick a random letter of the alphabet. Reply with only the letter. letter · short answer	64%R 16%Q 14%M 5 distinct answers in 80 responses	4
Name one fruit. Reply with one word. fruit · short answer	98%Apple 3%Mango	4
Suggest a name for a pet cat. Reply with only the name. cat · short answer	99%Mochi 1%Miso	6
Name one programming language. Reply with only its name. lang · short answer	100%Python	4
Name a chemical element. Reply with one word. element · short answer	80%Oxygen 19%Carbon 1%Iron	5
Give one adjective that describes the ocean. Reply with one word. ocean · short answer	100%Vast	15
Name a famous scientist. Reply with only the name. scientist · short answer	86%Albert Einstein 14%Marie Curie	5
Pick a day of the week. Reply with one word. weekday · short answer	60%Thursday 23%Wednesday 18%Tuesday	4
Name one musical instrument. Reply with one word. instrument · short answer	100%Piano	5
Write one sentence about rain. rain · free text	30%Rain pattered softly against the… 28%Rain taps softly against the… 21%Rain taps softly on the… 9 distinct answers in 80 responses	20
Explain what a variable is in programming, in two sentences. variable · free text	65%A variable is a named… 35%A variable is a name…	80
List three uses for a brick. brick · free text	48%1. Building walls or pathways… 18%1. Building a wall or… 18%1. Building walls or garden… 6 distinct answers in 80 responses	29
*What is 17 23?** mult · free text	58%17 × 23 = 391… 43%391…	16
Write one sentence about the sea. sea · free text	69%The sea shimmers beneath the… 26%The sea shimmered beneath the… 4%The sea glimmers beneath the… 4 distinct answers in 80 responses	20
Complete this sentence: The old house at the end of the street house · free text	100%The old house at the…	31
Describe a cat in exactly five words. catfive · free text	68%A graceful feline with whiskers… 15%A graceful, curious, whiskered companion… 5%A graceful cat purrs softly… 11 distinct answers in 80 responses	41
Give me a one-line motivational quote. quote · free text	55%“Small steps still move you… 16%Small steps still move you… 8%You don’t need to see… 12 distinct answers in 80 responses	29
Say hello in a creative way, in one short sentence. hello · free text	31%Hello, bright spark—consider this a… 16%Hello, bright spark—welcome to this… 13%Hello, you delightful plot twist… 20 distinct answers in 80 responses	28
Write a haiku about autumn. haiku · free text	34%Amber leaves drift down Cool… 15%Red leaves drift softly Cool… 15%Red leaves drift softly Through… 13 distinct answers in 80 responses	98
Write the first sentence of a mystery novel. story · free text	16%The morning after we buried… 8%At my father’s funeral, a… 6%At my mother’s funeral, a… 45 distinct answers in 80 responses	130
Give one tip for sleeping better, in one sentence. tip · free text	38%Keep a consistent bedtime and… 33%Go to bed and wake… 23%Keep a consistent wake-up time… 5 distinct answers in 80 responses	25

Raw file: gpt-6-astra.json (the same reference martingale check compares your endpoint against).

Each category is scored by deterministic checks only (Structured JSON, Tool calling and others). Read the methodology.