OpenAI's October changelog, backfilled: GPT-6 Rosalind, ChatGPT subscription API credits, and 12-hour prompt caching

OpenAI's official API changelog is the first source for this site's timeline, but developers.openai.com had returned 403 to us since 2026-09-01 — for two months we kept updates alive through official SDK release notes, GitHub announcements, and the news RSS, marking feature-class changelog items unverified. On 2026-11-01 the domain became reachable again (HTTP 200), and the official changelog, pricing page, model cards, and guide pages' .md variants all fetched successfully the same day. This article puts the most important September-October feature entries on record from the changelog text, each with its source and fetch date.

1. GPT-6 Rosalind: the life-sciences model (10-30)

The changelog's Oct 30 entry, verbatim:

Released GPT-6 Rosalind, bringing specialized biological reasoning with scientific tool use to complex life sciences research. Access is available to organizations approved for Rosalind Research through the trusted-access program for internal research by authorized users.

The pricing page's Life sciences models section (fetched 2026-11-01) lists three tiers:

gpt-6-rosalind also has long-context rates ($30 input / $112.50 output) and 30-minute / 12-hour cache write prices ($18.75 / $37.50); both the research and discovery variants have empty long-context and cache-write columns. Access, per the pricing page: Access is limited to approved internal research through the trusted-access program, and eligible organizations keep getting access to the latest GPT-Rosalind models as they are released. As of 2026-11-01 the model catalog has no standalone Rosalind model card (models/gpt-6-rosalind returns 404) and the help.openai.com program page is unreachable to this site — the verifiable boundary is the changelog entry and this pricing table.

2. ChatGPT subscriptions now grant API credits: up to $500 a month (10-28)

The changelog's Oct 28 entry: Eligible ChatGPT Pro and Business subscribers can now claim up to $500 per month in free API credits as a subscription benefit. The official guide (guides/chatgpt-api-credits, fetched 2026-11-01) gives the full table:

The rules (guide wording): Business credits are pooled and capped at $500 per workspace per month; annually-paid Business workspaces receive the full year in one issuance, expiring at the end of the annual billing cycle; the first grant is prorated for the time left in the current cycle; credits arrive automatically at renewal, expire at cycle end, and do not roll over; the API works with active credits without an API payment method. Eligibility: an active paid Pro or Business subscription at least seven days old (free trials and complimentary subscriptions do not qualify), the claimant owning both the subscription and the linked API organization, and the same email address on both. The guide also confirms a previously unrecorded fact: ChatGPT Pro comes in $100 / $200 / $500 tiers.

3. GPT-6.1 Terra released into the API (10-26)

The changelog's Oct 26 entry: Released GPT-6.1 Terra, a fast, efficient model for visual reasoning, precise computer use, and clear, natural writing. The model card and full pricing are in this batch's companion piece — GPT-6.1 Terra lands in the API: the model card, official pricing, and the GPT-6 family price table: a 1.05M context window, 922K max input, $1/$5 short context, 5% cached reads, and a six-level effort ladder including none.

4. 12-hour prompt caching (10-20)

The changelog's Oct 20 entry, verbatim: You can now choose 12-hour prompt caching in the API. 12-hour prompt caching keeps shared instructions, tools, and context cached longer to benefit from lower cached-input pricing, even when requests are hours apart. 12-hour cache writes cost 2.5x the input token price.

The official guide (fetched 2026-11-01) puts the tier table on the request-level prompt_cache_options.ttl:

Read rates are the same in both tiers: 0.1x input for most models, 0.05x for GPT-6.1 Sol and GPT-6.1 Terra. The guide's economics (at the 0.1x read rate): one write plus one full reuse costs 1.35x the ordinary input cost, versus 2x for processing it twice without caching; across ten requests, one write and nine reads cost 2.15x, versus 10x uncached. Mechanics: the request-level ttl applies to the implicit breakpoint and to explicit breakpoints without their own ttl; prompt_cache_breakpoint.ttl overrides it for an individual breakpoint; omitting both preserves default 30m behavior; unsupported values return 400 invalid_request_error; a single request can mix tiers, and each prefix token bills at the longest lifetime covering it.

{
  "model": "gpt-6.1-sol",
  "prompt_cache_options": { "mode": "explicit", "ttl": "30m" }
}

One dated status: on the SDK side (openai-python v3.42.0 tag, checked 2026-11-01) the request-level ttl docstring still reads 30m is currently the only supported value — the official guide has documented 12h first, and the SDK type docstring has not caught up. This site's previous SDK article asserted that the request level could not take 12h based on that day's SDK types; it now carries an update note pointing here.

5. Ultrafast expands three times, with official pricing

Ultrafast went from limited preview to three product surfaces (changelog wording):

The pricing page's Ultrafast section (fetched 2026-11-01) lists exactly two models: gpt-6-astra at $60/$300 short context and gpt-6.1-sol at $12/$60 — both 6x standard; long context astra $120/$450, 6.1-sol $24/$90. Ultrafast has separate rate limits, and higher limits require contacting an account team (guide wording). Note that GPT-5.6 Sol is not in the current Ultrafast pricing table — its 08-13 wording (14x, limited preview) is a historical entry outside this table; as of 2026-11-01 the pricing page is the governing source.

6. Tier reforms and service-tier controls

• Usage tiers five to three (10-06): Simplified API usage tiers from five to three: Build, Launch, and Grow. Organizations automatically upgrade as total credit purchases reach tier minimums.
• Service-tier project controls (10-21): You can now choose which service tiers projects can use in Platform settings. Organization settings can restrict service tiers for all projects.

7. The Agents API and GPT-Live cadence, September into October

Agents API: public beta 09-10 (a managed Codex harness, durable sessions, hosted or self-hosted sandboxes); 09-29 computer use in an OpenAI-hosted browser (website access approvals and sign-in handled by your application); 10-09 hosted sandboxes became pre-warmable, and sandbox compute is now billed during active turns — no more charges for idle time between turns; 10-16 sandboxes include the Artifact SDK and skills for spreadsheets, presentations, and Word documents.

GPT-Live: 10-16 opened prompt-to-voice to all API customers — create a custom voice from a text description; voice creation is free and speech generation bills as regular GPT-Live sessions — and added nine natural voices: German Nuri / Juni / Flitz, French Brise / Sillage, Hindi Nira, Japanese Harema / Shida, Korean Noeul. The pricing page's GPT-Live sessions section now lists gpt-live-1 sessions at $0.05 per minute, billed per second without rounding up (backend model and tool usage charged separately) — where this site previously had to mark Live API pricing as unannounced, this is the official answer; the model card page is reachable the same day (audio and text in and out, Jul 31, 2025 knowledge cutoff).

8. New pricing-page appearances: cyber, a legal configuration, and the FedRAMP uplift

The 2026-11-01 pricing page has three blocks this site had not recorded before:

• Cyber models section (labeled our latest Daybreak models): gpt-5.6-cyber and gpt-5.5-cyber, short context $12.50/$75 each (cached $1.25), long-context columns empty. The changelog's Aug 7 entry records the Daybreak Blue / Red access tiers (model tags include gpt-5.6-cyber and gpt-daybreak-*-latest) for approved defenders — the entries previously seen hidden in models.json now have a public pricing page. They are a different generation from gpt-5.4-cyber, removed from the API on 2026-10-01.
• Legal model configurations section: gpt-6-astra-law $10/$50 (cached $1.00), with a note that GPT-6 Astra for Law promotional pricing runs at least through 2026-12-31.
• Two footer notes: regional processing endpoints carry a 10% uplift for models released on or after 2026-03-05; FedRAMP endpoints get the same 10% uplift starting 2026-11-01.

9. September backfill and earlier entries

• 09-29: released GPT-6.1 Sol ($2 input / $0.10 cached / $2.50 cache write / $10 output short context; multi-agent beta — the model can delegate work to subagents within a Responses request); astra Ultrafast (see section 5); Agents API computer use (see section 7).
• 09-25: fixed an image-encoding bug that had degraded image understanding in GPT-6 Sol / Luna; official advice is to rerun evaluations and retry affected workflows.
• 09-22: released GPT-6 Sol ($2/$10, cached $0.20) and GPT-6 Luna ($0.10/$0.50, cached $0.01).
• 09-15 / 09-10: API key governance in two steps — organization/project restrictions on new key types (service-account-only, user-owned-project-only, or none at all); expiration dates at creation with administrator-enforced maximum lifetimes.
• 07-30: the GPT-5.6 price cut — Luna down 80%, Terra down 20%; the same day, Priority processing was renamed Fast mode. This site's selection guides have been corrected in this batch to Terra $2/$12 and Luna $0.20/$1.20 per the official pricing page.
• 10-05 / 10-07: HIPAA compliance support moved into the product (eligible org admins can accept the standard BAA in Organization settings); the chat-latest snapshot was updated (OpenAI continues to recommend the GPT-6 family for production).

10. Same-day news RSS entries (10-30, article pages 403 — not covered standalone)

The official news RSS carried three announcements on 2026-10-30; the article pages return 403 to this site and their text cannot be verified, so this site does not cover them standalone and registers them by RSS title and one-line description: Making it easier for websites to express AI training preferences (description: OpenAI will honor a shared robots.txt signal to help website operators communicate their AI training preferences to participating providers); Community safety report and new protections for teens; Fair use benefits everyone. Date anchoring: RSS fetched 2026-11-01.

11. Next steps

• The Terra model card and family price table: GPT-6.1 Terra lands in the API: the model card, official pricing, and the GPT-6 family price table.
• This batch's SDK news (agent compaction/review settings, external keys, SIP DTMF): openai-python 3.42.0 and openai-node 7.46/7.47: agent compaction and review settings, external keys, SIP DTMF.
• The previous batch's breakpoint TTL semantics (with this batch's update note): openai-python 3.40/3.41 and openai-node 7.45.1: breakpoint TTL semantics, tool_usage, and Realtime item references.
• The site-wide model timeline (living changelog): OpenAI Models Release Notes (2026, Living Document).