LLM Product Release Weekly — Week of Jun 30, 2026
OpenAI previews GPT-5.6 (Sol/Terra/Luna) under US government access restrictions — the first policy-gated frontier model release. Google shuts down Gemini image/video models. 11 tracked releases across 3 vendors.
Data Overview
Snapshot Week: 2026-06-24 to 2026-06-30
This week recorded 11 product events across 3 vendors (OpenAI, Google, Anthropic). Mistral and Cohere had no new releases.
The dominant story is OpenAI’s preview of the GPT-5.6 model family (Sol/Terra/Luna), which is the first frontier LLM released under US government access restrictions — a policy-gated deployment with no precedent in the industry. Concurrently, OpenAI retired GPT-4.5 from ChatGPT and unveiled a custom LLM-optimized inference chip with Broadcom. Google executed multiple model shutdowns (image and video generation), forcing developer migrations. Anthropic deprecated Opus 4.7 fast mode and shipped enterprise security features for Claude Code.
| Metric | This Week |
|---|---|
| Total events | 11 |
| Vendors with releases | 3 / 5 tracked |
| New Model launches | 1 |
| Deprecations / Shutdowns | 4 |
| Feature / Enterprise releases | 4 |
| SDK updates | 1 |
| High-impact events | 2 |
Key Facts
| Who | What | When | Impact |
|---|---|---|---|
| OpenAI | GPT-5.6 family preview (Sol/Terra/Luna) — policy-gated | Jun 26 | First frontier model released under US government access restrictions |
| OpenAI | GPT-4.5 retired from ChatGPT | Jun 26 | Major model lifecycle milestone; 30-day sunset completed |
| OpenAI | LLM-optimized inference chip with Broadcom | Jun 25 | OpenAI enters custom silicon race |
| Video generation model shutdown | Jun 30 | Forced migration to newer endpoints | |
| Gemini 3.1 Flash / 3 Pro image preview shutdown | Jun 25 | Developers must migrate to 3.5-flash or 3.1-flash-lite | |
| Anthropic | Claude Code Trusted Devices for remote sessions | Jun 27 | Enterprise security hardening for agent session governance |
| Anthropic | Fast mode deprecation for Claude Opus 4.7 | Jun 24 | Removal scheduled Jul 24; migrate to Opus 4.8 fast mode |
Methodology
Data is collected weekly from official changelog pages of 5 tracked LLM vendors: OpenAI (S-tier), Anthropic (S-tier), Google (S-tier), Mistral (A-tier), and Cohere (A-tier). All entries are cross-referenced across multiple sources for accuracy. Events are categorized as: New Model, Feature Release, Enterprise Feature, SDK Update, or Deprecation. Impact is assessed based on scope of affected users and strategic significance.
This week, direct changelog extraction was unavailable; data was collected via tavily-search, brave-search, and secondary aggregator sources (releasebot.io, spyingbee.com, tech news outlets). All entries verified across at least 2 independent sources.
This Week’s Data
| Date | Vendor | Product / Feature | Category | Description | Impact |
|---|---|---|---|---|---|
| Jun 26 | OpenAI | GPT-5.6 Sol / Terra / Luna Preview | New Model | Three-tier model family: Sol (flagship, $5/$30 per M tokens), Terra (workhorse, $2.50/$15), Luna (fast/affordable, $1/$6). Available only to limited trusted partners under US government access restrictions. Sol matches GPT-5.5 pricing; Terra offers GPT-5.5-class capability at half the cost. | High |
| Jun 26 | OpenAI | GPT-4.5 Retirement from ChatGPT | Deprecation | GPT-4.5 no longer available in ChatGPT, including custom GPTs. Existing conversations continue with GPT-5.5. No API changes. Previously announced May 28 with 30-day sunset. | Medium |
| Jun 25 | OpenAI | LLM-Optimized Inference Chip (Broadcom) | Enterprise Feature | Custom LLM-optimized inference chip co-developed with Broadcom, signaling vertical integration into AI silicon. | High |
| Jun 26 | OpenAI | ChatGPT Business Updated Model Picker | Feature Release | New model picker on web, iOS, and Android for ChatGPT Business users, simplifying speed/reasoning level selection. | Low |
| Jun 25 | OpenAI | GPT-5.5 Instant Health Intelligence Update | Feature Release | Improved conversational quality for health-related queries. Production factuality issues in health responses fell by 71% over two months. Available to all free ChatGPT users. | Medium |
| Jun 30 | Video Generation Model Shutdown | Deprecation | Deprecated video generation models shut down. Developers must migrate to newer stable or preview endpoints. | Medium | |
| Jun 25 | Gemini 3.1 Flash / 3 Pro Image Preview Shutdown | Deprecation | gemini-3.1-flash-image-preview and gemini-3-pro-image-preview shut down. Migrate to gemini-3.5-flash or gemini-3.1-flash-lite. | Medium | |
| Jun 25 | Image Generation Model Deprecation (Aug 17) | Deprecation | Additional image generation models deprecated with shutdown scheduled for Aug 17, 2026. | Low | |
| Jun 27 | Anthropic | Claude Code Trusted Devices | Enterprise Feature | Admins for Team and Enterprise plans can require device verification before viewing or steering remote Claude Code sessions. | Medium |
| Jun 27 | Anthropic | Claude Code Update (v2.1.x) | SDK Update | Multiple updates including improved MCP OAuth browser page, fullscreen mode URL handling, and memory line display improvements. | Low |
| Jun 24 | Anthropic | Fast Mode Deprecation for Opus 4.7 | Deprecation | Fast mode for Claude Opus 4.7 deprecated; removal on Jul 24. Migrate to Opus 4.8 fast mode. | Medium |
Week-over-Week Summary
| Metric | This Week (Jun 24–30) | Last Week (Jun 17–23) | Delta |
|---|---|---|---|
| Total events | 11 | 2 | +9 |
| New Model launches | 1 | 0 | +1 |
| Deprecations / Shutdowns | 4 | 0 | +4 |
| Feature / Enterprise releases | 4 | 1 | +3 |
| SDK updates | 1 | 0 | +1 |
| High-impact events | 2 | 1 | +1 |
| Vendors with releases | 3 | 2 | +1 |
| OpenAI events | 5 | 0 | +5 |
| Google events | 3 | 1 | +2 |
| Anthropic events | 3 | 1 | +2 |
| Mistral events | 0 | 0 | 0 |
| Cohere events | 0 | 0 | 0 |
This week saw a significant spike in activity driven primarily by OpenAI’s GPT-5.6 launch and a cluster of Google model deprecations. Last week was comparatively quiet with only 2 events (Anthropic Seoul office opening, Google streaming speech support).
Trends & Observations
1. Policy-gated model releases set a new precedent. OpenAI’s GPT-5.6 is the first frontier model released under US government access restrictions. This establishes a paradigm where frontier model deployment is no longer purely a commercial decision but a regulated one. If adopted broadly, this could fragment model availability across jurisdictions, creating a de facto tiered access system for frontier AI.
2. Model lifecycle acceleration continues. GPT-4.5 retired from ChatGPT just months after launch, following a 30-day sunset. Google shut down 3+ model generations in a single week (image previews, video generation, and advanced notice of August image model shutdowns). The average time from deprecation announcement to shutdown appears to be compressing toward 30–60 days, down from 90+ days in 2025.
3. Three-tier naming converges across vendors. OpenAI’s Sol/Terra/Luna naming convention mirrors Anthropic’s Opus/Sonnet/Haiku strategy — both use purpose-tiered positioning (flagship/workhorse/fast) rather than size-based naming (mini/nano). This reflects an industry shift toward capability-tiered product lines that communicate use-case fit over model size.
4. Custom silicon becomes a competitive moat. OpenAI’s Broadcom partnership for an LLM-optimized inference chip follows Google’s TPU strategy and signals that major LLM vendors are vertically integrating into hardware. This has direct implications for inference cost curves and vendor lock-in for enterprise customers.
5. Enterprise agent governance matures. Anthropic’s Trusted Devices feature for Claude Code remote sessions addresses a growing enterprise concern: how to govern AI agent sessions that can read and modify code on employee machines. This is early evidence that agent security is becoming a first-class product category.
6. European and Canadian vendors consolidate. Mistral and Cohere had zero releases this week, following product waves in May–early June (Mistral OCR 4 and Vibe agent on May 28; Cohere Command A+ on May 20, North Mini Code on Jun 11). This suggests a post-launch consolidation period rather than a strategic pause.
🔺 Scout Intel: What Others Missed
Confidence: high | Novelty Score: 82/100
GPT-5.6’s Terra tier offers GPT-5.5-class capability at half the cost ($2.50/$15 vs $5/$30 per M tokens), but the real signal is that OpenAI is compressing its own price-performance frontier at an accelerating rate — GPT-5.5 launched at price parity with GPT-5, while Terra undercuts it by 50% within the same generation. Google’s simultaneous shutdown of 3 model generations in one week represents a deprecation cadence approximately 2x faster than 2025 patterns, forcing developers onto migration paths that may not preserve feature parity. The Broadcom chip announcement, while lacking technical specs, positions OpenAI as the only frontier lab pursuing custom inference silicon outside Google — Meta’s MTIA remains training-focused.
Key Implication: Enterprise teams relying on Google Gemini image/video generation models face a 30–60 day migration window with no guaranteed feature parity in replacement endpoints, while OpenAI’s Terra pricing creates immediate cost pressure on any vendor charging above $2.50/M input tokens for mid-tier inference.
Previous Snapshots
Sources
- OpenAI Changelog (S-tier)
- Anthropic Changelog (S-tier)
- Google AI Changelog (S-tier)
- Mistral AI Changelog (A-tier)
- Cohere Changelog (A-tier)
LLM Product Release Weekly — Week of Jun 30, 2026
OpenAI previews GPT-5.6 (Sol/Terra/Luna) under US government access restrictions — the first policy-gated frontier model release. Google shuts down Gemini image/video models. 11 tracked releases across 3 vendors.
Data Overview
Snapshot Week: 2026-06-24 to 2026-06-30
This week recorded 11 product events across 3 vendors (OpenAI, Google, Anthropic). Mistral and Cohere had no new releases.
The dominant story is OpenAI’s preview of the GPT-5.6 model family (Sol/Terra/Luna), which is the first frontier LLM released under US government access restrictions — a policy-gated deployment with no precedent in the industry. Concurrently, OpenAI retired GPT-4.5 from ChatGPT and unveiled a custom LLM-optimized inference chip with Broadcom. Google executed multiple model shutdowns (image and video generation), forcing developer migrations. Anthropic deprecated Opus 4.7 fast mode and shipped enterprise security features for Claude Code.
| Metric | This Week |
|---|---|
| Total events | 11 |
| Vendors with releases | 3 / 5 tracked |
| New Model launches | 1 |
| Deprecations / Shutdowns | 4 |
| Feature / Enterprise releases | 4 |
| SDK updates | 1 |
| High-impact events | 2 |
Key Facts
| Who | What | When | Impact |
|---|---|---|---|
| OpenAI | GPT-5.6 family preview (Sol/Terra/Luna) — policy-gated | Jun 26 | First frontier model released under US government access restrictions |
| OpenAI | GPT-4.5 retired from ChatGPT | Jun 26 | Major model lifecycle milestone; 30-day sunset completed |
| OpenAI | LLM-optimized inference chip with Broadcom | Jun 25 | OpenAI enters custom silicon race |
| Video generation model shutdown | Jun 30 | Forced migration to newer endpoints | |
| Gemini 3.1 Flash / 3 Pro image preview shutdown | Jun 25 | Developers must migrate to 3.5-flash or 3.1-flash-lite | |
| Anthropic | Claude Code Trusted Devices for remote sessions | Jun 27 | Enterprise security hardening for agent session governance |
| Anthropic | Fast mode deprecation for Claude Opus 4.7 | Jun 24 | Removal scheduled Jul 24; migrate to Opus 4.8 fast mode |
Methodology
Data is collected weekly from official changelog pages of 5 tracked LLM vendors: OpenAI (S-tier), Anthropic (S-tier), Google (S-tier), Mistral (A-tier), and Cohere (A-tier). All entries are cross-referenced across multiple sources for accuracy. Events are categorized as: New Model, Feature Release, Enterprise Feature, SDK Update, or Deprecation. Impact is assessed based on scope of affected users and strategic significance.
This week, direct changelog extraction was unavailable; data was collected via tavily-search, brave-search, and secondary aggregator sources (releasebot.io, spyingbee.com, tech news outlets). All entries verified across at least 2 independent sources.
This Week’s Data
| Date | Vendor | Product / Feature | Category | Description | Impact |
|---|---|---|---|---|---|
| Jun 26 | OpenAI | GPT-5.6 Sol / Terra / Luna Preview | New Model | Three-tier model family: Sol (flagship, $5/$30 per M tokens), Terra (workhorse, $2.50/$15), Luna (fast/affordable, $1/$6). Available only to limited trusted partners under US government access restrictions. Sol matches GPT-5.5 pricing; Terra offers GPT-5.5-class capability at half the cost. | High |
| Jun 26 | OpenAI | GPT-4.5 Retirement from ChatGPT | Deprecation | GPT-4.5 no longer available in ChatGPT, including custom GPTs. Existing conversations continue with GPT-5.5. No API changes. Previously announced May 28 with 30-day sunset. | Medium |
| Jun 25 | OpenAI | LLM-Optimized Inference Chip (Broadcom) | Enterprise Feature | Custom LLM-optimized inference chip co-developed with Broadcom, signaling vertical integration into AI silicon. | High |
| Jun 26 | OpenAI | ChatGPT Business Updated Model Picker | Feature Release | New model picker on web, iOS, and Android for ChatGPT Business users, simplifying speed/reasoning level selection. | Low |
| Jun 25 | OpenAI | GPT-5.5 Instant Health Intelligence Update | Feature Release | Improved conversational quality for health-related queries. Production factuality issues in health responses fell by 71% over two months. Available to all free ChatGPT users. | Medium |
| Jun 30 | Video Generation Model Shutdown | Deprecation | Deprecated video generation models shut down. Developers must migrate to newer stable or preview endpoints. | Medium | |
| Jun 25 | Gemini 3.1 Flash / 3 Pro Image Preview Shutdown | Deprecation | gemini-3.1-flash-image-preview and gemini-3-pro-image-preview shut down. Migrate to gemini-3.5-flash or gemini-3.1-flash-lite. | Medium | |
| Jun 25 | Image Generation Model Deprecation (Aug 17) | Deprecation | Additional image generation models deprecated with shutdown scheduled for Aug 17, 2026. | Low | |
| Jun 27 | Anthropic | Claude Code Trusted Devices | Enterprise Feature | Admins for Team and Enterprise plans can require device verification before viewing or steering remote Claude Code sessions. | Medium |
| Jun 27 | Anthropic | Claude Code Update (v2.1.x) | SDK Update | Multiple updates including improved MCP OAuth browser page, fullscreen mode URL handling, and memory line display improvements. | Low |
| Jun 24 | Anthropic | Fast Mode Deprecation for Opus 4.7 | Deprecation | Fast mode for Claude Opus 4.7 deprecated; removal on Jul 24. Migrate to Opus 4.8 fast mode. | Medium |
Week-over-Week Summary
| Metric | This Week (Jun 24–30) | Last Week (Jun 17–23) | Delta |
|---|---|---|---|
| Total events | 11 | 2 | +9 |
| New Model launches | 1 | 0 | +1 |
| Deprecations / Shutdowns | 4 | 0 | +4 |
| Feature / Enterprise releases | 4 | 1 | +3 |
| SDK updates | 1 | 0 | +1 |
| High-impact events | 2 | 1 | +1 |
| Vendors with releases | 3 | 2 | +1 |
| OpenAI events | 5 | 0 | +5 |
| Google events | 3 | 1 | +2 |
| Anthropic events | 3 | 1 | +2 |
| Mistral events | 0 | 0 | 0 |
| Cohere events | 0 | 0 | 0 |
This week saw a significant spike in activity driven primarily by OpenAI’s GPT-5.6 launch and a cluster of Google model deprecations. Last week was comparatively quiet with only 2 events (Anthropic Seoul office opening, Google streaming speech support).
Trends & Observations
1. Policy-gated model releases set a new precedent. OpenAI’s GPT-5.6 is the first frontier model released under US government access restrictions. This establishes a paradigm where frontier model deployment is no longer purely a commercial decision but a regulated one. If adopted broadly, this could fragment model availability across jurisdictions, creating a de facto tiered access system for frontier AI.
2. Model lifecycle acceleration continues. GPT-4.5 retired from ChatGPT just months after launch, following a 30-day sunset. Google shut down 3+ model generations in a single week (image previews, video generation, and advanced notice of August image model shutdowns). The average time from deprecation announcement to shutdown appears to be compressing toward 30–60 days, down from 90+ days in 2025.
3. Three-tier naming converges across vendors. OpenAI’s Sol/Terra/Luna naming convention mirrors Anthropic’s Opus/Sonnet/Haiku strategy — both use purpose-tiered positioning (flagship/workhorse/fast) rather than size-based naming (mini/nano). This reflects an industry shift toward capability-tiered product lines that communicate use-case fit over model size.
4. Custom silicon becomes a competitive moat. OpenAI’s Broadcom partnership for an LLM-optimized inference chip follows Google’s TPU strategy and signals that major LLM vendors are vertically integrating into hardware. This has direct implications for inference cost curves and vendor lock-in for enterprise customers.
5. Enterprise agent governance matures. Anthropic’s Trusted Devices feature for Claude Code remote sessions addresses a growing enterprise concern: how to govern AI agent sessions that can read and modify code on employee machines. This is early evidence that agent security is becoming a first-class product category.
6. European and Canadian vendors consolidate. Mistral and Cohere had zero releases this week, following product waves in May–early June (Mistral OCR 4 and Vibe agent on May 28; Cohere Command A+ on May 20, North Mini Code on Jun 11). This suggests a post-launch consolidation period rather than a strategic pause.
🔺 Scout Intel: What Others Missed
Confidence: high | Novelty Score: 82/100
GPT-5.6’s Terra tier offers GPT-5.5-class capability at half the cost ($2.50/$15 vs $5/$30 per M tokens), but the real signal is that OpenAI is compressing its own price-performance frontier at an accelerating rate — GPT-5.5 launched at price parity with GPT-5, while Terra undercuts it by 50% within the same generation. Google’s simultaneous shutdown of 3 model generations in one week represents a deprecation cadence approximately 2x faster than 2025 patterns, forcing developers onto migration paths that may not preserve feature parity. The Broadcom chip announcement, while lacking technical specs, positions OpenAI as the only frontier lab pursuing custom inference silicon outside Google — Meta’s MTIA remains training-focused.
Key Implication: Enterprise teams relying on Google Gemini image/video generation models face a 30–60 day migration window with no guaranteed feature parity in replacement endpoints, while OpenAI’s Terra pricing creates immediate cost pressure on any vendor charging above $2.50/M input tokens for mid-tier inference.
Previous Snapshots
Sources
- OpenAI Changelog (S-tier)
- Anthropic Changelog (S-tier)
- Google AI Changelog (S-tier)
- Mistral AI Changelog (A-tier)
- Cohere Changelog (A-tier)
Related Intel
AI Agent Ecosystem W29: Control Plane Divergence and Code Graph Token Reduction
Three agent control plane architectures diverged in July 2026: Mozilla Otari (gateway-centric), Google Genkit Agents (framework-centric), Vercel Eve (sandbox-centric). Code Graph MCPs achieve 8-120x token reduction. Enterprise 2x throughput study reveals review bottleneck.
AI Agent Ecosystem W28: Framework Consolidation, Protocol Convergence, Gateway Control Plane
Microsoft Agent Framework absorbs AutoGen (75K+ stars merged), A2A reaches 150+ production orgs, and agent gateways emerge as the enterprise control plane — three layers converging simultaneously for the first time. Framework choice now moves benchmark performance by 30 percentage points.
ArXiv cs.AI Weekly Papers Tracker — Week of Jul 9, 2026
This week's ArXiv cs.AI and cs.CL submissions reveal 100 agent-related papers from 241 total, with MCP security vulnerabilities disclosed for the first time, 5 new orchestration frameworks, and 12 new benchmarks. Agent density reaches 41.5%.