DROPGoogle launches Gemini 3.8 Flash TTS and Flash-Lite TTS///DROPGemini Omni brings 1080p generation to Google Vids///BREAKClaude Opus 5.5 is here: 1M context, $4 input / $20 output///BREAKOpenAI adds GPT-6 Sol and Luna to the GPT-6 family///PAPEROpus 5.5's system card documents gains—and evaluation limits///BREAKGrok 4.7 launches at $2 input / $6 output per million tokens///DROPGrok Voice Transcribe 2.0 targets difficult real-world audio///TRENDGoogle steers new Gemini API projects to 3.8 Flash or 3.5 Flash-Lite///DROPGoogle launches Gemini 3.8 Flash TTS and Flash-Lite TTS///DROPGemini Omni brings 1080p generation to Google Vids///BREAKClaude Opus 5.5 is here: 1M context, $4 input / $20 output///BREAKOpenAI adds GPT-6 Sol and Luna to the GPT-6 family///PAPEROpus 5.5's system card documents gains—and evaluation limits///BREAKGrok 4.7 launches at $2 input / $6 output per million tokens///DROPGrok Voice Transcribe 2.0 targets difficult real-world audio///TRENDGoogle steers new Gemini API projects to 3.8 Flash or 3.5 Flash-Lite///
DARC.DEV

[ 01 — Source-checked signal ]

AI news, with receipts

Real release dates. Original announcements. Clear availability and pricing caveats. This is an editorial snapshot—not an automatically updating newswire.

Source-checked Dated editorial snapshot · official sources · no invented resultsSources & corrections
DROPPRIMARY SOURCE

Google launches Gemini 3.8 Flash TTS and Flash-Lite TTS

Google's 23 September release adds voice design, consent-verified voice replication and two-speaker direction. Flash TTS targets creative fidelity; Flash-Lite targets high-volume generation. Both are rolling out through Gemini API and AI Studio.

Read Google

Editorial interpretation · not a measured result

Separate availability from the roadmap: voice remixing is described as coming soon, not shipping today.

Announcement date above · reviewed 23 Sep 2026

Practical next steps →
DROPPRIMARY SOURCE

Gemini Omni brings 1080p generation to Google Vids

Google says anyone with a Google Account can create 1080p video in Vids with Gemini Omni. The update adds scene extension and clip-duration controls. Flash-Lite TTS narration is still marked coming soon in this announcement.

Read Google

Editorial interpretation · not a measured result

A useful creator update, but a product rollout is not evidence that one video model wins every shot.

Announcement date above · reviewed 23 Sep 2026

Practical next steps →
BREAKPRIMARY SOURCE

Claude Opus 5.5 is here: 1M context, $4 input / $20 output

Released 22 September, Opus 5.5 replaces Opus 5 as the newest Opus. Standard token prices fall 20%; cache reads drop to $0.20/M. Anthropic reports 40% lower typical workload cost. Thinking is always on; Sonnet 5.5 and Haiku 5.5 are not yet released.

Read Anthropic

Editorial interpretation · not a measured result

The migration is more than a model-name edit: review always-on thinking, preserved context and safeguard fallback behavior.

Announcement date above · reviewed 23 Sep 2026

Practical next steps →
BREAKPRIMARY SOURCE

OpenAI adds GPT-6 Sol and Luna to the GPT-6 family

The 22 September API changelog lists GPT-6 Sol and GPT-6 Luna. Standard prices for prompts up to 272K input tokens are $2/$10 and $0.10/$0.50 per million input/output tokens respectively. Both accept text and images and return text.

Read OpenAI

Editorial interpretation · not a measured result

Astra, Sol and Luna are different tiers. Compare completed-task quality and the long-context surcharge, not just the headline rate.

Announcement date above · reviewed 23 Sep 2026

Practical next steps →
PAPERPRIMARY SOURCE

Opus 5.5's system card documents gains—and evaluation limits

The new system card covers agentic coding, computer use, alignment evaluations and deployment safeguards. Anthropic also reports weaknesses, including malicious instructions pasted into user prompts. Reported scores are not guarantees of safety in your application.

Read Anthropic

Editorial interpretation · not a measured result

Read the caveats alongside the headline benchmarks. A more robust model does not replace tool isolation and approval gates.

Announcement date above · reviewed 23 Sep 2026

Practical next steps →
BREAKPRIMARY SOURCE

Grok 4.7 launches at $2 input / $6 output per million tokens

xAI's 21 September release is available through the Grok API, Grok Build and Cursor. The current catalog lists 500K context. xAI reports 46.3% on CursorBench 4.0 and 37.6% on Terminal-Bench 4.0 at xhigh effort.

Read xAI / SpaceXAI

Editorial interpretation · not a measured result

These are provider-reported results. Keep reasoning effort, harness and budget attached when comparing them with other models.

Announcement date above · reviewed 23 Sep 2026

Practical next steps →
DROPPRIMARY SOURCE

Grok Voice Transcribe 2.0 targets difficult real-world audio

The 18 September release focuses on telephony, overlapping voices, accents and spoken credentials. xAI says it improves accuracy over Transcribe 1.0 at the same price, and links both public and internal evaluations.

Read xAI / SpaceXAI

Editorial interpretation · not a measured result

Measure word error rate on your own calls—including names and numbers—rather than treating clean-audio scores as production evidence.

Announcement date above · reviewed 23 Sep 2026

Practical next steps →
TRENDPRIMARY SOURCE

Google steers new Gemini API projects to 3.8 Flash or 3.5 Flash-Lite

Google's 18 September changelog limits Gemini 2.5 access to existing active users. The notice explicitly says these models are not deprecated and remain served; it recommends 3.8 Flash or 3.5 Flash-Lite for new projects.

Read Google

Editorial interpretation · not a measured result

Access restrictions, deprecation and retirement are different states. A trustworthy model catalog should not collapse them into 'dead'.

Announcement date above · reviewed 23 Sep 2026

Practical next steps →
DROPPRIMARY SOURCE

Gemini Notebook expands source-grounded study workflows

Google's 15 September announcement describes notebook conversations, mobile recording, interactive learning overviews and additional quiz formats. Availability differs by feature and account; the article marks several capabilities as staged rollouts rather than universally available.

Read Google

Editorial interpretation · not a measured result

Curate the source set and verify important citations. A grounded study artifact inherits both the strengths and the weaknesses of the documents you supply.

Announcement date above · reviewed 23 Sep 2026

Practical next steps →
DROPPRIMARY SOURCE

Gemini 3.8 Live can talk while tools work in the background

Google released Gemini 3.8 Live and Live Extended Thinking on 15 September. The Live API supports asynchronous function calling; Extended Thinking adds background reasoning while the conversation continues. Product access differs by surface.

Read Google

Editorial interpretation · not a measured result

Non-blocking tools change the voice UX. Test interruptions and completed-action confirmations, not just conversational fluency.

Announcement date above · reviewed 23 Sep 2026

Practical next steps →
DROPPRIMARY SOURCE

DeepSeek V4.1 Flash adds native vision and a new architecture

The 10 September release is served as deepseek-flash with 1M context. The pricing page lists $0.30/$1.20 peak and $0.15/$0.60 off-peak input/output rates per million tokens. The corrected changelog says V4 Pro continues after 14 September.

Read DeepSeek

Editorial interpretation · not a measured result

The later changelog corrects the original phase-out announcement. Use current service documentation when planning migration.

Announcement date above · reviewed 23 Sep 2026

Practical next steps →
DROPPRIMARY SOURCE

OpenAI's Agents API enters public beta

The 10 September API changelog introduces durable sessions, a managed Codex harness, context compaction and recovery. Developers can use hosted sandboxes or connect supported infrastructure and their own tools.

Read OpenAI

Editorial interpretation · not a measured result

Managed orchestration reduces glue code; it does not remove the need for permission boundaries and application-level evals.

Announcement date above · reviewed 23 Sep 2026

Practical next steps →
DROPPRIMARY SOURCE

GPT Image 2.5 Sunburst and Flare arrive in the API

OpenAI's 8 September changelog distinguishes Sunburst for editing precision from Flare for fast everyday image generation. Both are documented for the Image API and the Responses image-generation tool.

Read OpenAI

Editorial interpretation · not a measured result

Separate the image model from the model orchestrating the tool. The product name is not the endpoint contract.

Announcement date above · reviewed 23 Sep 2026

Practical next steps →
BREAKPRIMARY SOURCE

GPT-6 Astra starts the new GPT-6 generation

OpenAI introduced Astra on 3 September for coding, computer use, research and professional artifacts. Its current API model page lists a 1.05M total context window and $10/$50 standard input/output pricing, with higher rates above 272K input tokens.

Read OpenAI

Editorial interpretation · not a measured result

Tool-enabled integrations must use the Responses API. Review parameter changes and safety controls before swapping model IDs.

Announcement date above · reviewed 23 Sep 2026

Practical next steps →
DROPPRIMARY SOURCE

Gemini 3.8 Flash ships; Flash Cyber has separate access rules

Google's 2 September release makes 3.8 Flash generally available. The introductory $0.75/$3.75 rates run through 31 December 2026. Flash Cyber is provided to trusted defenders through the Fairwind Program—not unrestricted general access.

Read Google

Editorial interpretation · not a measured result

Record pricing expiry and access class alongside version numbers. 'Latest' is not a sufficient deployment specification.

Announcement date above · reviewed 23 Sep 2026

Practical next steps →
PAPERPRIMARY SOURCE

The current MCP specification changes protocol and authorization assumptions

The 2026-07-28 release introduces a stateless protocol core, multi-round-trip requests, routing headers, cacheable list results and authorization hardening. It retains compatibility paths while moving toward Client ID Metadata Documents for client registration.

Read Model Context Protocol

Editorial interpretation · not a measured result

This is a July specification, not a new September launch. Pin your SDK and protocol versions before adapting an older integration tutorial.

Announcement date above · reviewed 23 Sep 2026

Practical next steps →