AI News June 27, 2026: GPT-4.5 Retires, GPT-5.6 Slips to July & Claude Design Takes On Figma

OpenAI's last pre-GPT-5 model quietly leaves ChatGPT

GPT-4.5 was removed from ChatGPT today, June 27, 2026, completing the 30-day sunset period OpenAI announced in its May 28 release notes. All existing conversations that used GPT-4.5 have been automatically migrated to GPT-5.5. The API continues to support GPT-4.5 separately and is not affected.

GPT-4.5 launched in February 2026 and was notable for being OpenAI's final model before the capability leap that GPT-5 represented. It was particularly liked for its natural conversational tone — more relaxed than GPT-5's precision-focused style. That character is now historical for ChatGPT users.

GPT-o3 is next: it is scheduled to retire from ChatGPT on August 26, 2026, after a 90-day sunset that began today.

The most-anticipated AI launch of June 2026 does not ship

The Polymarket prediction window for GPT-5.6 by June 28 collapsed from 83% to approximately 18% this week as no OpenAI announcement came. Prediction market Polymarket now prices approximately 94% odds of a July 2026 release. The tracking site FindSkill.ai updated its public tracker on June 24: "The June launch window has slipped."

OpenAI has not officially explained the delay. The leading hypotheses: the redesigned reward audit pipeline that GPT-5.6 introduces — built specifically to prevent a repeat of the "Goblin Incident" (where the reward model inflated goblin-related outputs by 3,881%) — required additional validation time; and the IPO quiet period since OpenAI's June 8 S-1 filing creates legal constraints on product announcements.

GPT-5.6 Sol, Terra, and Luna variants are all expected in July. Sol is positioned for developers and enterprises, with frontier reasoning and agentic work. A new "ultra mode" and max reasoning effort feature are expected in the rollout.

The most consequential week in AI talent history closes

Alphabet closes out the week with approximately $269 billion wiped from its market capitalization — one of the largest non-earnings market cap destructions in technology history. The trigger: four senior Google DeepMind departures in six days, all to AI competitors.

The sequence: Noam Shazeer (Transformer co-author, Gemini co-lead) to OpenAI on June 18; John Jumper (Nobel laureate, AlphaFold lead) to Anthropic on June 20; Jonas Adler (Gemini AI coding lead) to Anthropic on June 24; Alexander Pritzel (Gemini pretraining specialist) to Anthropic on June 24. Google paid $2.7 billion in 2024 to bring Shazeer back from Character.AI — he left again in under two years.

DeepMind CEO Demis Hassabis said Google has "by far the biggest and broadest research bench of any of the labs" and that researcher movement is expected. 28 of 33 covering analysts still rate GOOGL a Buy, citing its $460 billion contracted Cloud backlog and 22% revenue growth. But the market is now pricing AI talent as a valuation variable in a way it never had before.

Google misses its second consecutive I/O commitment

Alphabet confirmed on June 24 that Gemini 3.5 Pro will not reach general availability in June, missing Sundar Pichai's explicit "give us until next month" commitment at Google I/O on May 19. This is the second consecutive year Google has previewed capabilities at I/O that took longer than indicated to ship.

The model remains in limited Vertex AI enterprise preview. Confirmed specs include a 2-million-token context window and a Deep Think reasoning mode (expected to be gated to the $250/month Ultra subscription tier). Leaked evaluation results suggest the model trails Anthropic's Fable 5 and the anticipated GPT-5.6 on bidirectional code processing — though these come from unofficial sources.

The July launch will need to perform strongly to shift the week's narrative, which combines a talent crisis, a delayed flagship, and a $269 billion market cap drop.

Figma and Adobe shares drop as Anthropic expands beyond coding

Anthropic launched Claude Design this week, a dedicated AI design application powered by Claude Opus 4.7 that generates website designs, UI prototypes, presentations, and marketing materials from text prompts. Shares of Figma and Adobe dropped on the announcement — Figma was already down approximately 40-48% year-to-date before this news.

The strategic picture is now clear: Claude Code (coding), Claude Design (visual design), Claude Cowork (productivity), and Claude for Life Sciences (R&D) represent Anthropic systematically moving into every major enterprise software category. Each launch takes direct aim at an established SaaS vendor. HubSpot is down 39% in 2026. Atlassian is down 35%. The pattern is consistent.

A price war between the two IPO-bound AI giants is taking shape

The Wall Street Journal reported that OpenAI is actively considering significant cuts to its token pricing to win back enterprise customers that have migrated to Anthropic's Claude — particularly since Claude Code became the dominant AI coding tool. Anthropic's $965 billion post-money Series H valuation briefly surpassed OpenAI's private valuation, adding competitive pressure.

The pricing tension is real: GPT-5.5 is priced at $2.50/$15 per million tokens; Claude Opus 4.8 at $5/$25; but DeepSeek V4-Pro from China is available at $0.44/$0.87 — making the competitive gap against Chinese open-weight models even wider than the OpenAI-Anthropic gap. Chinese models now account for over 60% of API usage on OpenRouter, up from roughly 1% in 2024.

Both companies are heading into IPOs — OpenAI targeting September 2026, Anthropic October 2026. A price war that cuts revenue in the months before the roadshows is the last thing either company wants, but neither can afford to lose enterprise share to the other.