OpenAI Ecosystem Week 43 Briefing (2026-09-01): mTLS GA, Sora 2 Countdown, Shutdown Calendar

The OpenAI ecosystem week 43 briefing covers official announcements as of 2026-09-01. No new models this week -- the main line is API infrastructure: security (mTLS), cost observability (the Prompt Caching dashboard), and speed (Ultrafast mode) all landed within days of each other. Meanwhile the Sora 2 shutdown countdown entered its final 23 days, and the official deprecations page contains a full fall/winter shutdown calendar that every team should walk through line by line. The OpenAI ecosystem briefing is a sourced summary of key ecosystem events, with official references for every entry.

1. The Main Line: Three Infrastructure Releases

1.1 mTLS and X.509 Workload Identity Are GA (8-29)

The official changelog entry for 8-29: Mutual TLS (mTLS) and X.509 workload identity federation are now generally available for the OpenAI API. The configuration entry point is the Platform console -- certificates and X.509 identity providers can be configured directly there, with access controlled by your organization's roles and permissions.

For most individual developers this entry is easy to skip. But if you are building enterprise-grade integrations -- especially under zero-trust networking, service identity auditing, or compliance requirements -- OpenAI now officially offers a certificate-grade identity path for API callers: instead of a bare API key, your calls to the OpenAI API can be authenticated by a certificate that is rotatable, revocable, and bound to a workload identity.

1.2 The Prompt Caching Dashboard Ships (8-20)

Prompt caching itself has existed for a while (repeated prefixes hitting the cache save money); what shipped this time is observability. The official Prompt Caching dashboard shows your cache hit rate over time, cache reads per write, and the breakdown of cache-read, cache-write, and uncached tokens. Metrics are filterable by model and service tier.

Practical advice: if you run long system prompts with multi-turn conversations or batch evaluation workloads, check the hit rate first. A common cause of a low hit rate is an unstable prefix (timestamps or random values placed early in the prompt). The token breakdown makes this class of problem easy to locate.

1.3 Ultrafast Mode Still in Flight (8-13 Recap)

Ultrafast mode, announced on 8-13, is a new API service tier exclusive to GPT-5.6 Sol, officially described as up to 14x faster than Standard processing. It is currently in limited preview for select customers, with a sign-up channel for updates. Of this week's three infrastructure items it is the only one not yet broadly available -- noted here for status, pending a rollout announcement.

2. Sora 2 Enters the Shutdown Countdown (9-24, 23 Days Out)

The deprecations page entry announced on 2026-03-24 lists the full set of shutdown targets and the date: the Videos API, sora-2, sora-2-pro, sora-2-2025-10-06, sora-2-2025-12-08, and sora-2-pro-2025-10-06, all shutting down on 2026-09-24.

Two details deserve special attention:

1. No official replacement is suggested. In the deprecations table, the "Recommended replacement" column for this batch is "---" -- unlike the transcription models (explicitly pointed at gpt-live-transcribe / gpt-transcribe) or the legacy GPT snapshots (explicitly pointed at the GPT-5.6 family). For the video generation direction, OpenAI currently has no official answer.
2. The entire API line is shutting down, not just model snapshots -- the Videos API itself is on the list. Production code still calling it cannot be fixed by swapping a model name; it needs a full migration assessment.

There are 23 days until 9-24. Teams still using the video generation API should schedule "migrate or decommission" this week -- do not count on a delay; no extension has been announced so far.

3. The Fall/Winter Shutdown Calendar (2026-09 to 2027-02)

Sorting every unexpired entry on the official deprecations page by shutdown date produces the calendar below. Each row maps directly to an announcement batch on that page:

Two "shutdown-like" dates also belong on your team calendar: 2026-11-12, when OpenAI's termination of model supply to Cursor takes effect for individual users (enterprise transition until 2026-12-31), and 2026-11-21, the official floor date of the GPT-5.6 Sol promotional price of $4/$20 (after that date, pricing is whatever the official pricing page says).

4. Early-August Backfill: Four Easy-to-Miss Updates

• 8-7 · Daybreak tiers: for approved defenders, Daybreak now offers two tiers. Daybreak Blue covers most defensive security work, using general-purpose models such as GPT-5.6 Sol for vulnerability discovery, secure code review, detection engineering, incident response, malware analysis, and patch validation. Daybreak Red requires separate approval and opens purpose-trained models such as GPT-5.6 Cyber for authorized vulnerability reproduction, exploit validation, penetration testing, and red teaming.
• 8-6 · chat-latest snapshot update: chat-latest points to the latest model available in ChatGPT for Plus and Pro users, and the underlying snapshot is updated regularly; OpenAI continues to recommend GPT-5.6 Sol for production API usage.
• 8-5 · Fast mode long context: all three GPT-5.6 tiers can now run long-context prompts exceeding 272K tokens in Fast mode, up to 2.5x faster than the Standard tier.
• 8-4 · API-key dimension in usage dashboards: the Usage and Costs dashboards support filtering and grouping by API key, and the Usage API and Costs API support the same dimension for programmatic reporting and attribution.

5. This Week's Action List

1. Video API users: audit your codebase for Videos API / sora-2 calls and migrate or decommission before 9-24 (no official replacement -- assess your own path).
2. Teams running long prompts: open the Prompt Caching dashboard, check hit rate and token breakdown, and fix unstable prefixes first.
3. Enterprise integration teams: evaluate whether mTLS / X.509 belongs in your next security review -- as of 8-29 this is an officially supported capability.
4. Everyone: copy the shutdown calendar in section 3 into your team calendar, especially the 10-23 and 11-30 batch dates.

6. Next Steps

• For the full methodology on shutdowns and migrations, see this site's companion piece published today: OpenAI Deprecation Lifecycle Guide: Notice Periods, Shutdown Calendar, Migration Checklist.
• Last week's briefing (Assistants API shutdown / Sol price cut / transcription deprecations): OpenAI Ecosystem Week 42 Flash: Assistants API Shut Down, Sol Price Cut, Transcription Deprecations.
• Model selection and the full GPT-5.6 family picture: The complete guide to GPT models (2026-07): GPT-5.6 Sol, Terra, Luna.