Six Lifecycle Moves in 21 Days: o3 Retired, DALL-E GPT Gone, Claude Sonnet 5 Discounts End Today
Six AI product lifecycle events landed between August 10 and August 31, 2026. Taken individually each is routine maintenance. Together they mark one of the more compressed model-lifecycle resets since frontier labs began accelerating release cadences.
The full ledger:
| Date | Vendor | Event |
|---|---|---|
| Aug 13 | DeepSeek | V4 Pro 0813 promoted to general availability, no press release |
| Aug 17 | Imagen 4 API endpoint shut down | |
| Aug 26 | OpenAI | o3 removed from ChatGPT model picker |
| Aug 28 | Tencent | Hy4 Preview weights released under Apache 2.0 |
| Aug 30 | OpenAI | DALL-E GPT removed from ChatGPT model picker |
| Aug 31 | Anthropic | Claude Sonnet 5 promotional pricing expires |
o3 and DALL-E GPT Exiting ChatGPT
OpenAI flagged the DALL-E GPT retirement in ChatGPT release notes on July 31 and executed on August 30. The dedicated entry point is gone; image generation continues through ChatGPT Images and GPT-builder tools with image generation enabled. The capability was not removed, only the named model picker entry.
o3 was cleared from the model picker on August 26. It had already been superseded by the o-series variants shipping with GPT-5.4 and GPT-5.5 thinking tiers, but it maintained a picker slot through the summer. Both removals follow a pattern: OpenAI now announces retirement dates weeks in advance, then executes them without additional ceremony.
DeepSeek V4 Pro Goes GA Silently
DeepSeek pushed V4 Pro 0813 to general availability on August 13 without a blog post. The -0813 build is the current production checkpoint, MIT license, weights available. On Terminal-Bench 2.1, independent third-party evaluators have placed V4 Pro at 87.9 — one-tenth of a point behind Claude Fable 5’s 88.0 on the same benchmark, and roughly 2.9 points ahead of Claude Opus 4.8.
The silent GA is consistent with DeepSeek’s behavior since V4-Flash in July. No formal announcement, changelog updated, weights uploaded. The model shows up in leaderboard evaluations before most of the industry notices it has shipped.
Tencent Hy4 Preview: 770B MoE, Apache 2.0
Tencent released Hy4 Preview weights on August 28 under Apache 2.0. The model is a 770B total-parameter mixture-of-experts architecture activating 49B parameters per token. Terminal-Bench 2.1 puts it at 85.4% resolve rate per independent evaluation posted the same week. The open release continues Tencent’s pattern with Hy3 and Hy3-Preview: ship weights under permissive license alongside the hosted API.
Apache 2.0 on a 770B MoE is significant for enterprise adoption. Teams that cannot send data to a hosted API can now run Hy4 Preview on-premises with full commercial rights.
Claude Sonnet 5 Promotional Pricing Ends Today
Anthropic ran discounted pricing on Claude Sonnet 5 for several weeks as a developer adoption driver. That promotional rate expires August 31. Teams that built cost projections around the discounted rate need to rerun their numbers before today’s switch. Standard Sonnet 5 pricing applies from September 1 onward.
The timing — one day after OpenAI cleared DALL-E GPT — is coincidental but concentrates two separate pricing recalculations for developers watching both vendors into a 48-hour window.
Pattern
Each of these events has a practical reason: models get superseded, APIs consolidate, promotional windows close on schedule. What is unusual about August 2026 is the density. Labs are running faster release cycles, which means faster deprecation cycles, which means developers are managing more simultaneous end-of-life windows than at any prior point in the frontier model era.
The practical implication: pin model IDs in production, not product names. And check for promotional pricing expiry dates before committing cost models to a planning cycle.