AI Model Launch Tracker

What we captured

OpenAI announced GPT‑Live, a new generation of full‑duplex voice models (GPT‑Live‑1 and GPT‑Live‑1 mini) powering ChatGPT Voice, launched to ChatGPT users globally on July 8, 2026; OpenAI plans to bring the models to the API soon and at launch uses GPT‑5.5 in the background.

• • OpenAI announced GPT‑Live on July 8, 2026 as a new generation of voice models powering ChatGPT Voice.
• • GPT‑Live is built on a full‑duplex architecture and can listen and speak at the same time.
• • At launch GPT‑Live will use GPT‑5.5 in the background; OpenAI will update the background frontier model over time.
• • OpenAI is beginning to roll out GPT‑Live‑1 and GPT‑Live‑1 mini to ChatGPT users globally and plans to bring them to the API.
• • Update July 31, 2026: OpenAI added SynthID watermarking for supported audio, a public verification tool for provenance signals, and API access for verification.

Source evidence

Confidence 95% / Completeness 87%

What we captured

Poolside released Laguna XS 2.1, an open-weight 33B-A3B mixture-of-experts model for agentic coding and local deployment.

• • 33B total parameters with 3B activated parameters per token.
• • SWE-bench Multilingual score of 63.1%, up 5.4 points from Laguna XS.2.
• • Available in FP8, INT4, and NVFP4 quantized checkpoints.
• • Poolside reported doubled achieved tokens per second using DFlash speculator models.
• • Paid pricing is $0.10 input, $0.20 output, and $0.05 cache-read per 1M tokens.

Source evidence

Confidence 100% / Completeness 100%

What we captured

Anthropic launched Claude Sonnet 5 as a generally available agentic model across Claude plans and the Claude API, priced at $2 per million input tokens and $10 per million output tokens.

• • Anthropic announced Claude Sonnet 5 on June 30, 2026.
• • The model is available across Free, Pro, Max, Team, and Enterprise plans and through the Claude API.
• • Permanent pricing is $2 per million input tokens and $10 per million output tokens.
• • Anthropic reported improved agentic performance and lower overall undesirable behavior rates compared with Sonnet 4.6.

Source evidence

Confidence 100% / Completeness 92%

What we captured

Anthropic publicly released Claude Fable 5 (a Mythos-class model) with topic-based safeguards; available via API and Enterprise with published per-million-token prices.

• • Article states Anthropic "publicly released Claude Fable 5, its first 'Mythos-class' model".
• • Fable 5 is publicly available while Mythos 5 remains restricted to a "small group of cyberdefenders" vetted through Project Glasswing.
• • Anthropic built topic-based safeguards in Fable 5 that funnel cybersecurity/biology/chemistry queries to Opus 4.8 and warn users; biology/chemistry classifier now applies to all such queries in Fable 5.
• • Anthropic says Fable 5 will be available to API and Enterprise users at $10 per million input tokens and $50 per million output tokens "starting today."
• • Subscription plans include access to Fable 5 through June 22, after which users must purchase "usage credits" to access the model.

Source evidence

What we captured

Poolside launched Laguna XS.2 and Laguna M.1 into preview for agentic coding. Laguna XS.2 is available as Apache 2.0 open weights, while both models are available through Poolside’s API and OpenRouter for a limited time.

• • Poolside announced the release of Laguna XS.2 and Laguna M.1 on April 28, 2026.
• • Laguna XS.2 has 33B total parameters and 3B active parameters.
• • Laguna XS.2 weights are available on Hugging Face under Apache 2.0.
• • Laguna M.1 has 225B total parameters and 23B active parameters.
• • The May 26 update states that both models are served at 256K context lengths and that Laguna M.1 open weights are available.

Source evidence

Confidence 99% / Completeness 91%

What we captured

Anthropic announced Claude Mythos Preview on April 7, 2026; the system card reports a 1,000,000-token context window, pricing listed as $25/$125 per million tokens, and a restricted-access rollout.

• • Article: "Anthropic announced Claude Mythos Preview on April 7, 2026."
• • Article: "The system card reports ... a 1M-token context window."
• • Article: "Mythos is priced at $25/$125 per million tokens — 5x above Opus ..."
• • Article: "The restricted-access rollout is consistent with a deeper pattern ..."
• • Article: "Mythos is Anthropic’s newest frontier model, internally codenamed 'Capybara.'"

Source evidence

What we captured

OpenAI announced a new ChatGPT Images experience and the GPT Image 1.5 image-generation model, rolling out December 16, 2025 in ChatGPT and available in the API.

• • Article published December 16, 2025 announcing the release.
• • "Today, we’re releasing a new version of ChatGPT Images... The new Images model is rolling out today in ChatGPT for all users, and is available in the API as GPT Image 1.5."
• • ChatGPT Images "makes precise edits while keeping details intact, and generates images up to 4x faster."
• • "Image inputs and outputs are now 20% cheaper in GPT Image 1.5 as compared to GPT Image 1."
• • The new Images experience in ChatGPT is "rolling out today for most users, with Business and Enterprise access coming later."
• • GPT Image 1.5 "delivers all the same improvements as ChatGPT Images: it’s stronger at image preservation and editing than GPT Image 1."

Source evidence

What we captured

OpenAI announced GPT-5.2, a new model series for professional knowledge work, with rollout in ChatGPT for paid plans and immediate availability in the API.

• • OpenAI published 'Introducing GPT‑5.2' on December 11, 2025.
• • Article states: 'In ChatGPT, GPT‑5.2 Instant, Thinking, and Pro will begin rolling out today, starting with paid plans.'
• • Article states: 'In the API, they are available now to all developers.'
• • Article notes GPT‑5.2 achieves near-100% accuracy on the 4-needle MRCR variant out to 256k tokens.
• • Article defines '256k represents 256 * 1,024 = 262,144 tokens.'
• • OpenAI describes GPT‑5.2 as 'the most capable model series yet for professional knowledge work' and highlights improvements in spreadsheets, presentations, code, vision, long-context understanding, and tool use.

Source evidence

What we captured

OpenAI describes GPT‑4b micro, an experimental GPT model trained for protein engineering and developed for research purposes (not broadly available); the model can handle prompts up to 64,000 tokens and was used with Retro Biosciences to design enhanced protein variants.

• • Source (Aug 22, 2025): 'we collaborated with the Applied AI team at Retro Bio... to create and research the impact of GPT‑4b micro, a miniature version of GPT‑4o specialized for protein engineering.'
• • Source: 'GPT‑4b micro was developed for research purposes and is not broadly available.'
• • Source: 'we could run prompts as large as 64,000 tokens during inference' (article states this context size was used during development/inference).
• • Source: OpenAI reports GPT‑4b micro was used with Retro Biosciences to design RetroSOX and RetroKLF variants that in vitro 'achieved greater than a 50-fold higher expression of stem cell reprogramming markers than wild-type controls.'
• • Article publication date: 2025-08-22.

Source evidence

What we captured

Inflection AI introduced Inflection 3.0 as the foundation of Inflection for Enterprise and made its large language model available through a new commercial API.

• • Inflection 3.0 powers Inflection for Enterprise.
• • Inflection AI released a commercial API for the model at developers.inflection.ai.
• • The enterprise system integrates with Intel Gaudi 3 AI accelerators and Intel Tiber AI Cloud.

Source evidence

Confidence 95% / Completeness 87%