OpenAI's 2026 from the platform side: the GPT-5.x release-and-retire cadence, the Sora shutdown, API changes, and where the company stands mid-year.
On the model side: GPT-5.5 serves as the working default, GPT-5.6 is rolling out, and GPT-5.4 was retired on July 23 β a two-quarter release-and-retire rhythm that now defines the platform. The year's most consequential move was subtraction, not addition: Sora was pulled on March 25, 2026, and its API sunsets on September 24, 2026, a rare full retreat from a headline product category.
This page tracks OpenAI's platform and company moves. For the ChatGPT app itself, see ChatGPT updates.
GPT-5.6 reached GA on July 9, 2026, after a limited preview that began June 26. It shipped as three explicitly named tiers rather than one router-mediated model:
| Model | Role | Input / Output per 1M |
|---|---|---|
| Sol | Flagship | $5 / $30 |
| Terra | Balanced, everyday work | $2.50 / $15 |
| Luna | Most cost-efficient | $1 / $6 |
On July 30, 2026 OpenAI cut Luna by 80% and Terra by 20% β three weeks after GA. An intra-quarter cut that large on a just-launched family is a competitive response, not a promotion: Gemini 3.6 Flash sells frontier-adjacent capability at $1.50/$7.50, and Claude Sonnet 5 opened at $2/$10.
GPT-5 tried to hide tier selection behind a real-time router in 2025 and users read it as being silently downgraded. Three named tiers with published prices is the direct product consequence β and the clearest evidence that the 2025 launch lesson stuck.
| Model | Status (July 2026) |
|---|---|
| GPT-5 (Aug 2025 launch) | Superseded β see our archived GPT-5 review |
| GPT-5.3 Codex family | Tooling absorbed into 5.4+ |
| GPT-5.4 / 5.4-mini | Retired July 23, 2026 |
| GPT-5.5 Instant | Current default |
| GPT-5.6 | Staged rollout in progress |
For API builders the operational lesson is now explicit: pin capabilities, not model names. Deprecation windows have shortened to months, and OpenAI's own guidance pushes automatic model-routing over hard-coded IDs.
The video model that defined 2024β2025's text-to-video race was pulled from service on March 25, 2026; the API follows on September 24, 2026. Whatever mix of compute economics, moderation burden and strategic refocus drove it, the practical outcome is clear: production video workloads migrated to Veo 3.1, Runway Gen-4.5 and Kling, and "OpenAI = every modality" stopped being a safe planning assumption.
Reported figures put Sora at roughly $15M/day in operating cost against ~$2.1M in total lifetime revenue, at about $1.30 per 10-second clip in raw generation cost. This wasn't a capability retreat β it was a unit-economics failure, and it's the most instructive data point of 2026 for anyone modeling generative-media costs.
The September 24 API sunset is hard, not a soft deprecation. Migrate now β the migration guide covers where each workload lands. Doing it in August rather than September buys you the slack to pick the right destination rather than the fastest one.
OpenAI spent 2026 defending rather than extending its lead: Anthropic overtook it in private-market valuation with a $965B Series H in May, Google's Gemini 3.x line closed the capability gap at the cost-efficiency end, and the Windsurf acquisition collapse (contractual complications with Microsoft) handed the agentic-IDE market to Cognition. None of this dents OpenAI's consumer mindshare β ChatGPT remains the default word for "AI" for hundreds of millions β but the frontier is now contested on every axis: capability, price, and enterprise trust.
Consumer distribution and ecosystem. ChatGPT remains the default word for "AI" for hundreds of millions of people, and no competitor matches its combination of voice, images, agents and third-party integrations.
Price. Google's 3.6 Flash at $1.50/$7.50 and Anthropic's Sonnet 5 at introductory $2/$10 both undercut GPT-5.6 Sol by a wide margin β which is what the July 30 Luna and Terra cuts were answering.
Agentic coding. Claude Sonnet 5 posts 63.2% on SWE-bench Pro and 81.2% on OSWorld-Verified, with Opus 4.8 at 69.2% on the former. That's the segment Claude Code turned into a $2.5B run-rate business.
A headline modality. Sora's exit is the only case in 2026 of a frontier lab fully abandoning a product category it had defined.
Not decline β contest. OpenAI is competing on every axis simultaneously rather than setting the terms on any single one, which is a normal position for a market leader in a maturing category and a novel one for this company.
Sources: Releasebot OpenAI models tracker Β· OpenAI ChatGPT release notes Β· Spectrum AI Lab on the Sora shutdown Β· SolidAITech model guide Β· DevOps.com on the Windsurf deal collapse
Last updated: July 29, 2026
Anthropic's 2026 in business terms: the $65B Series H at a $965B valuation, confidential IPO filing, $47B run-rate, and enterprise adoption of Claude.
Every major ChatGPT update through July 2026: the GPT-5.5/5.6 transition, Dreaming V3 memory, Lockdown Mode, and what free vs paid gets you now.
Every major Claude update through July 2026: the Claude 5 family, Reflect dashboard, MCP upgrade, Claude Code milestones, pricing and access.
Copilot's 2026: Agent Mode general availability, agentic code review that ships fix PRs, the Copilot CLI, desktop app, and much faster indexing.
Every major Gemini update through July 2026: the 3.5/3.6 Flash generation, Gemini Omni multimodal creation, Nano Banana 2, and the agentic-era strategy.
Midjourney's 2026 in full: V8 alpha to the V8.2 default, native 2K without upscaling, 50% faster and 25% cheaper renders, and 21-second video.
Notion's 2026: autonomous Custom Agents, the credit billing switch on May 4, agents that read your files, 35β50% cheaper runs, and what it costs now.
Perplexity's browser year: Comet free on every platform since March, iPad and iOS upgrades, Deep Research generating decks, and enterprise controls.
