Advertise here

GPT-6.2 Astra and Sol tie for #1 in our tests

Published 2026-11-17 · Data captured 17 Nov 2026, 20:20 UTC

GPT-6.2 Sol at max effort and GPT-6.2 Astra at high effort both scored 9.99/10 and solved all 69 attempts, the first two places on the leaderboard. Sol max ranks #1 because its run cost less ($1.913 vs $2.232), but Astra high answered 2.7 times faster (10.8 s vs 28.9 s). Astra high beats GPT-6 Astra max by 0.1 points for 44% of the cost. GPT-6.2 holds six of the top seven places.

See current GPT-6.2 Astra results →

  • Tested through OpenRouter on the OpenAI endpoint on November 17, 2026, launch day. Prices per million tokens: Astra $8 input and $40 output, Sol $2 and $10. Prompts over 272k tokens cost more; our tests stay well below that.
  • Costs are recorded spend for one full run (23 tests, 3 repeats each). Failed API calls stay in the results and count as unsolved attempts.
  • Updated November 17, 2026: our pick for the most capable model is now GPT-6.2 Sol max instead of Astra high. The data is unchanged.

Key results

A tie at the top. Sol max and Astra high both scored 9.99 with 69 of 69 attempts correct. Astra xhigh and max reached the same score for more money.

Astra high is the faster pick. 10.8 s per answer, against 28.9 s for Sol max. Sol max costs 14% less per run.

A clear step over GPT-6 Astra. Astra high scored 0.1 points more than GPT-6 Astra max for 44% of its cost ($2.232 vs $5.044), and answered 5.7 times faster.

Sol xhigh is the value pick near the top. 9.93 for $1.091, #5 overall, 0.04 points above GPT-6.1 Sol max with 28% of its response time.

OpenAI's claims, checked. OpenAI says Astra costs much less per task than GPT-6 Astra, and that Sol nearly matches Astra on coding. In our suite, Astra high cost 44% of GPT-6 Astra max, and Sol matched Astra's best score at max effort.

Model Effort Rank Score Solved tests Correct attempts Run cost Cents per solved test Avg response
GPT-6.2 Sol max #1 9.99 23/23 69/69 $1.913 8.318 28.9 s
GPT-6.2 Astra high #2 9.99 23/23 69/69 $2.232 9.704 10.8 s
GPT-6.2 Sol xhigh #5 9.93 22/23 68/69 $1.091 4.959 17.4 s
Gemini 3.6 Flash high #6 9.91 22/23 68/69 $1.512 6.872 18.9 s
GPT-6.1 Sol max #10 9.89 22/23 66/69 $1.006 4.571 62.0 s
GPT-6 Astra max #11 9.89 22/23 66/69 $5.044 22.923 61.6 s
The top GPT-6.2 settings against the previous OpenAI leaders and Gemini 3.6 Flash high.

Score, cost and speed

The four top scores are close together, but cost and speed are not. GPT-6 Astra max is the most expensive and the slowest of this group.

Score (out of 10, higher is better)

GPT-6.2 Sol max 9.99
GPT-6.2 Astra high 9.99
GPT-6.2 Sol xhigh 9.93
Gemini 3.6 Flash high 9.91
GPT-6.1 Sol max 9.89
GPT-6 Astra max 9.89

Full-run cost in USD (lower is better)

GPT-6.2 Sol max $1.913
GPT-6.2 Astra high $2.232
GPT-6.2 Sol xhigh $1.091
Gemini 3.6 Flash high $1.512
GPT-6.1 Sol max $1.006
GPT-6 Astra max $5.044

Average response time in seconds (lower is better)

GPT-6.2 Sol max 28.9 s
GPT-6.2 Astra high 10.8 s
GPT-6.2 Sol xhigh 17.4 s
Gemini 3.6 Flash high 18.9 s
GPT-6.1 Sol max 62.0 s
GPT-6 Astra max 61.6 s

Which effort to use

Astra stops improving at high. Xhigh and max scored the same 9.99, but max cost 3.5 times as much as high and took 4.4 times as long per answer. Medium (9.91 for $1.632) is close behind high.

Sol needs more effort to reach the top. It goes from 8.44 at low to 9.85 at high, 9.93 at xhigh and 9.99 at max. Below high, Astra is the better choice.

Sol max recorded one provider error on test 12. We retried that attempt once and it answered correctly.

Model Effort Rank Score Solved tests Correct attempts Run cost Cents per solved test Avg response
GPT-6.2 Astra low #25 9.56 21/23 65/69 $1.433 6.820 5.6 s
GPT-6.2 Astra medium #7 9.91 22/23 68/69 $1.632 7.415 6.7 s
GPT-6.2 Astra high #2 9.99 23/23 69/69 $2.232 9.704 10.8 s
GPT-6.2 Astra xhigh #3 9.99 23/23 69/69 $3.262 14.182 18.8 s
GPT-6.2 Astra max #4 9.99 23/23 69/69 $7.741 33.654 47.3 s
GPT-6.2 Sol low #125 8.44 18/23 58/69 $0.354 1.963 4.4 s
GPT-6.2 Sol medium #35 9.45 20/23 64/69 $0.446 2.230 5.8 s
GPT-6.2 Sol high #15 9.85 21/23 67/69 $0.726 3.455 10.5 s
GPT-6.2 Sol xhigh #5 9.93 22/23 68/69 $1.091 4.959 17.4 s
GPT-6.2 Sol max #1 9.99 23/23 69/69 $1.913 8.318 28.9 s
Every tested effort level of GPT-6.2 Astra and Sol.

Strengths and weak points

Sol max and Astra high scored 10 in every category except data parsing (9.95), where every model in this group lost the same points.

Gemini 3.6 Flash, GPT-6.1 Sol and GPT-6 Astra lost most of their points in domain-specific questions. Sol xhigh lost them in instruction following.

GPT-6.2 Astra high

Agentic 10.00
Anti-AI Tricks 10.00
Coding 10.00
Combined 10.00
Data parsing and extraction 9.95
Domain specific 10.00
General Intelligence 10.00
Instructions following 10.00
Puzzle Solving 10.00
Tool Calling 10.00
Trivia 10.00

GPT-6.2 Sol max

Agentic 10.00
Anti-AI Tricks 10.00
Coding 10.00
Combined 10.00
Data parsing and extraction 9.95
Domain specific 10.00
General Intelligence 10.00
Instructions following 10.00
Puzzle Solving 10.00
Tool Calling 10.00
Trivia 10.00

GPT-6 Astra max

Agentic 10.00
Anti-AI Tricks 10.00
Coding 10.00
Combined 10.00
Data parsing and extraction 9.95
Domain specific 7.67
General Intelligence 10.00
Instructions following 10.00
Puzzle Solving 10.00
Tool Calling 10.00
Trivia 10.00

Gemini 3.6 Flash high

Agentic 10.00
Anti-AI Tricks 10.00
Coding 10.00
Combined 10.00
Data parsing and extraction 9.95
Domain specific 8.22
General Intelligence 10.00
Instructions following 10.00
Puzzle Solving 10.00
Tool Calling 10.00
Trivia 10.00
Category scores out of 10 (higher is better).

Who should choose which

GPT-6.2 Sol max: top score for the lowest cost. Our pick for the most capable model, if 29-second answers are fine.

GPT-6.2 Astra high: the same score with answers 2.7 times faster, for 17% more per run.

GPT-6.2 Sol xhigh or high: 9.93 for $1.091 or 9.85 for $0.726, a strong choice when cost matters.

Skip Astra xhigh and max for this kind of work: no higher score, much higher cost.

Model details

OpenAI released GPT-6.2 Astra and Sol on November 17, 2026 in the OpenAI API, ChatGPT Work and Codex. Astra is the flagship. Sol is the everyday model, priced the same as GPT-6.1 Sol.

Context window 1,050,000 tokens
Input text, image, file
Output text
Price per 1M tokens $8.000 input · $40.000 output · $0.400 cache read · $10.000 cache write
Reasoning Always on; effort low, medium, high, xhigh, max (default medium)
Parameters Not disclosed by OpenAI
Availability Closed. OpenAI API, ChatGPT Work and Codex since November 17, 2026
Prices observed 2026-11-17

Method and limits

Every model ran the same 23 tests three times through OpenRouter. Scores are out of 10, ranks are leaderboard positions at capture time, and comparisons use the 440 complete, non-archived chat results. Equal scores are ranked by cost per solved test.

GPT-6.2 was tested on launch day. GPT-6 Astra, GPT-6.1 Sol and Gemini 3.6 Flash were tested in late September and early October 2026. Tests 14, 15 and 22 kept the same prompts, answers and pass scores; only their recorded evaluator version differs.

The top scores are separated by fractions of a point, so small differences are not a decisive lead. Response times include provider and workload effects. Costs are recorded spend, not a price guarantee.

Sources

Linked model pages show current results, which can differ from these figures.