The newest GPT model is GPT-6.1 Sol, released on September 29, 2026. It shipped 25 days ago, and the line's usual gap is about 75 days.
How long it has been since the last release against the line's usual gap and its recent ones — the rhythm, not a date.
Latest: GPT-6.1 Sol · Sep 29, 2026
Every Monday, only when something shipped: the AI models that came out, what each one changed, and which lines are due next.
We email you a link to confirm first · unsubscribe anytime · read past issues · about the newsletter · RSS for the GPT line
Newest first — what each version changed versus the one before, in OpenAI's own numbers.
One week after GPT-6 Sol, it scores 7 points higher on computer-use tasks and 4 points higher on the all-round intelligence index at the same $2 / $10 price.
Compared with GPT-6 Sol · the model it replaces
Completes 7 in 10 tasks when operating a computer on its own
OSWorld 2.0 — using real desktop apps to finish tasks
71.4%
▲ up from 64.4% · +7 points · both at max effort
Scores 4 points higher on the all-round intelligence index
Artificial Analysis Intelligence Index v4.3.2 — one score averaging 10 hard tests
52
▲ up from 48 · +4 points · both at max effort
Costs the same as the model it replaces
Price per 1M tokens — what developers pay, input / output
$2 / $10
same as GPT-6 Sol · cached input halved to $0.10
Half the price of GPT-5.6 Sol with about half as many factual mistakes, while its all-round score rises only one point.
Compared with GPT-5.6 Sol · the model it replaces
Makes about half as many factual mistakes
OpenAI factuality test — errors in real-world conversations
about half
▲ vs GPT-5.6 Sol · OpenAI internal test
Scores 1 point higher on the all-round intelligence index
Artificial Analysis Intelligence Index v4.3.2 — one score averaging 10 hard tests
48
▲ up from 47 · +1 point
Costs half as much as the model it replaces
Price per 1M tokens — what developers pay, input / output
$2 / $10
▲ down from $4 / $20 · 50% cheaper
OpenAI's GPT-5.5 roughly doubles its score on a 1M-token long-document test and solves more command-line coding tasks than GPT-5.4 — but its API price doubles.
Compared with GPT-5.4 · the model it replaces
Roughly doubles its score at finding facts across book-length input
OpenAI MRCR v2 (8-needle, 512K-1M tokens) — finding and linking facts across book-length input
74.0%
▲ up from 36.6% · roughly doubled
Solves more command-line coding tasks
Terminal-Bench 2.0 — complex command-line coding workflows
82.7%
▲ up from 75.1% · +7.6 points
Costs developers twice as much to run
Price per 1M tokens — what developers pay, input / output
$5 / $30
price doubled — was $2.50 / $15
OpenAI's first general-purpose model with built-in computer use: it completes 75.0% of OSWorld-Verified desktop tasks versus 47.3% for GPT-5.2 and accepts up to 1M tokens of context, at a higher price per token.
Compared with GPT-5.2 · the model it replaces
Completes 3 in 4 tasks when operating a computer on its own
OSWorld-Verified — using real desktop apps to finish tasks
75.0%
▲ up from 47.3% · +27.7 points
Costs developers more per token than GPT-5.2
Price per 1M tokens — what developers pay, input / output
$2.50 / $15
was $1.75 / $14 · input up 43%, output up 7%
GPT-5.2 Thinking beats or ties top industry professionals in 70.9% of GDPval work-task comparisons, according to expert judges, and fixes 80.0% of SWE-bench Verified software bugs versus 76.3% for GPT-5.1, at a 40% higher API price per token.
Compared with GPT-5.1 · the model it replaces
Fixes 8 in 10 real software bugs
SWE-bench Verified — fixing real bugs in real software projects; GPT-5.2 Thinking and GPT-5.1 Thinking, each at its maximum API effort
80.0%
▲ up from 76.3% for GPT-5.1 · +3.7 points
Costs developers 40% more per token than GPT-5.1
Price per 1M tokens — what developers pay, input / output
$1.75 / $14
was $1.25 / $10 · input and output both up 40%
GPT-5 fixes 74.9% of real software bugs on SWE-bench Verified versus 69.1% for o3 and makes about 80% fewer factual errors on LongFact and FactScore prompts, with cheaper input but pricier output tokens.
Compared with o3 · the model it replaces
Fixes 3 in 4 real software bugs
SWE-bench Verified — fixing real bugs in real software projects
74.9%
▲ up from 69.1% · +5.8 points
Cheaper to send text in, pricier for the answers than o3
Price per 1M tokens — what developers pay, input / output
$1.25 / $10
was $2 / $8 · input 37.5% cheaper, output 25% pricier
o3 fixes 69.1% of real software bugs on SWE-bench Verified versus 48.9% for o1 and makes 20% fewer major errors on hard real-world tasks in external expert reviews, at a launch price a third lower.
Compared with o1 · the model it replaces
Fixes 7 in 10 real software bugs
SWE-bench Verified — fixing real bugs in real software projects
69.1%
▲ up from 48.9% · +20.2 points
Costs a third less than o1 at launch
Price per 1M tokens — what developers pay, input / output
$10 / $40
▲ down from $15 / $60 · 33% cheaper
o1-preview, the first model in OpenAI's new o1 series of reasoning models that spend more time thinking before they answer, solved 56.7% of AIME 2024 math problems versus 13.4% for GPT-4o; the full o1, then still in development, scored 83.3%.
Compared with GPT-4o · comparison chosen by OpenAI
o1-preview solves more than half the problems on a math olympiad qualifying exam
AIME 2024 — qualifying exam for the USA Math Olympiad; majority answer over 64 attempts; o1-preview is the version released that day
56.7%
▲ up from 13.4% for GPT-4o · +43.3 points · the full o1, then still in development, scored 83.3%
Costs developers 3 to 4 times as much as GPT-4o
Price per 1M tokens — what developers pay, input / output
$15 / $60
vs $5 / $15 for GPT-4o at launch · input 3x, output 4x
GPT-4o handles text, audio and images in one model, answers speech in 320 milliseconds on average (GPT-4 Voice Mode took 5.4 seconds) and costs half as much as GPT-4 Turbo in the API.
Compared with GPT-4 Turbo · the model it replaces
Answers spoken questions in about a third of a second
Voice response time — average delay before it answers spoken input
320 ms
▲ down from 5.4 seconds in GPT-4 Voice Mode
Costs half as much as GPT-4 Turbo
Price per 1M tokens — what developers pay, input / output
$5 / $15
▲ down from $10 / $30 · 50% cheaper
GPT-4 Turbo reads up to 128K tokens in one prompt, more than 300 pages of text, and costs a third as much as GPT-4 for input and half as much for output.
Compared with GPT-4 · the model it replaces
Reads more than 300 pages of text in a single prompt
Context window — how much text the model can read at once
128K tokens
▲ up from 8K tokens for GPT-4, or 32K for its larger version
Costs a third as much as GPT-4 for input and half as much for output
Price per 1M tokens — what developers pay, input / output
$10 / $30
▲ down from $30 / $60 for GPT-4 · 3x cheaper input, 2x cheaper output
GPT-4 passed a simulated bar exam with a score around the top 10% of test takers, where GPT-3.5 scored around the bottom 10%, and launched in ChatGPT Plus with API access by waitlist.
Compared with GPT-3.5 · the model it replaces
Scores around the top 10% of test takers on a simulated bar exam
Uniform Bar Exam, simulated — a US lawyer licensing exam; OpenAI's estimated percentile among human test takers
about 90th percentile
▲ score 298 / 400 · up from 213 / 400, about 10th percentile, for GPT-3.5
Costs developers 15 to 30 times as much as GPT-3.5 Turbo
Price per 1M tokens — what developers pay, input / output; GPT-4 with 8K context
$30 / $60
vs $2 / $2 for GPT-3.5 Turbo, the GPT-3.5 model used in ChatGPT · input 15x, output 30x