Independent · Source-cited · No sponsored rankingsSources linked in every article
AIToolRanked
AI News · 5 min read

OpenAI 8-Hour Window: 40% Off GPT-6.2 API, vs Batch and Flex

OpenAI's 8-hour Responses API window is 40% below Standard on GPT-6.2. How the 2-hour window, Batch, and Flex compare when a job can wait.

RA
· Founder, AIToolRanked
Flat night-sky illustration of a clock with a shaded eight-hour arc, a queue of request cards and three shrinking coin stacks beside a crescent moon
On this page

What OpenAI added on November 17

On November 17, 2026, OpenAI's API changelog ↗ added a public beta: "Choose a two-hour or eight-hour background completion window in the Responses API public beta. Model token rates are 30% lower than Standard for the two-hour window and 40% lower for the eight-hour window." The models on that entry are gpt-6.2-astra, gpt-6.2-sol, gpt-6.1-terra, and gpt-6-luna.

OpenAI Developers ↗ posted it on X on November 18, US time, with a follow-up ↗. The lines are "save up to 40% per token than Standard on GPT-6.2 Astra or GPT-6.2 Sol" and "choose a 2h or 8h target for the first response. These discounted windows are now in public beta." The example is sending meeting notes before you log off and asking for a morning brief. Sulat ↗ (JP Caparas, November 19) and S5 Labs ↗ covered the beta. The picture above is an original graphic created by aitoolranked. We have not sent a request on either window.

How to turn the discount on

The background mode guide ↗ is the switch. Set background to true and service_tier to default, which is Standard. With auto, the request qualifies only if the API selects Standard. Set background_options.completion_window to 2h or 8h. The default, soonest, has no discount. Leave stream unset, or set it to false.

The guide says the API "targets completion within the selected window after it accepts the request." That is a target, not a guarantee. OpenAI's own examples are a same-day ticket triage and an overnight document review.

Results come back on a response.completed webhook, or you poll the response ID. A continuation_id you choose "does not add prior history to the input," the API reference ↗ says. Reuse the window, that id, the model, and the project on follow-ups for a higher queue priority. The discount covers each eligible request, follow-ups included. History still needs previous_response_id, or a resend of the earlier turn if that response has expired.

The 2h and 8h jobs share the Batch/Background rate limit with Batch, and input tokens count at submission. Standard requests cannot reuse the prompt cache those jobs create, even with the same prompt. Flex is gone on the newest models. The guide says: "Starting with GPT-6.2 Astra and Sol, new models launch without Flex processing (service_tier: flex)." The GPT-6.2 guide ↗ points long Responses work here. Model context is in our November 17 note.

What GPT-6.2 costs if you can wait

Figures are per million tokens at short context. The pricing page ↗ defines that as 272,000 input tokens or fewer. Standard Astra is $8 in and $40 out. Sol is $2 and $10. Batch is half of Standard: Astra $4 and $20, Sol $1 and $5. Sulat describes Batch as a file of requests with turnaround within 24 hours. The window dollar amounts are our arithmetic, 30% and 40% off those Standard rates. Sulat prints the same prices.

OptionIn / out per 1MWait
Standard$8 / $40On request
2-hour window$5.60 / $28Target: 2 hours
8-hour window$4.80 / $24Target: 8 hours
Batch API$4 / $20Within 24 hours

The table is GPT-6.2 Astra. Sol is $1.40 and $7 at two hours, and $1.20 and $6 at eight hours.

On Terra and Luna, check Flex first. It is still listed at half of Standard: GPT-6.1 Terra $0.50 and $2.50 (Standard $1 and $5), GPT-6 Luna $0.05 and $0.25 (Standard $0.10 and $0.50). Our arithmetic puts eight-hour Terra at $0.60 and $3.00, and Luna at $0.06 and $0.30, so Flex is cheaper per token on both. The windows matter most on GPT-6.2, which has no Flex row.

The same page lists Ultrafast for Astra and Sol at 3 times Standard through February 28, 2027, and 6 times from March 1, 2027. That fast tier is separate from this discount.

What people are saying

  • "8h window on gpt-6.2 sol is fine for my nightly jobs. 40% off for work nobody is watching is an easy trade" — serx ↗ (@serxzsz)

  • "Batch is still cheaper at 50% off ($4/$20) if you can send it all up front. the window is for jobs that need tool calls mid-way" — Plausible ↗ (@plausible63)

  • "A job with 10M tokens in and 2M out on Astra: $160 at Standard, $96 in the 8h window." — Tengkolok ↗ (@Tengkolok24)

  • "The catch: your normal traffic can't reuse the prompt cache those requests create." — Tengkolok ↗ (@Tengkolok24)

  • "OpenAI just quietly dropped Flex for its newest models." — Awanio ↗ (@Awanioz)

  • "evals, nightly reports, backfills don't need answers in seconds." — Karthik Varma ↗ (@Karthikvarmamkv)

Reddit was not reachable for this story.

Our take

Use the window when the job can sit, and Batch when the file is ready. Nightly reports, evals, backfills, and agents that call a tool mid-run are the fit. Batch is still half off when the file can go up front. That is Plausible's split, and Tengkolok's $160 versus $96 matches the Astra table.

Check Flex before Terra or Luna. Half off beats a 40% window, and Flex is still listed for those two. Astra and Sol have no Flex row. Awanio's line matches the guide. Build for a late result: the window is a target, Standard traffic cannot reuse the cache these jobs write, and they share a rate limit with Batch. Google has its own batch discount; December 1 is in our Gemini API price note. For one key across labs, see Perplexity Orchestrator. Gemini 4 Neon is the other model note this week. Coding tools: Codex vs Claude Code and the 2026 coding guide.

Verdict: Eight hours for nightly and multi-step work on GPT-6.2. Batch when every request is ready now. Flex before Terra or Luna. Watch the shared Batch/Background limit and the cache split.

More from the news desk · About Rai Ansar