AI Model Launch Tracker

What we captured

DeepSeek announced an experimental vision model supporting Chat Completions, Messages, and Responses requests, with a 384-token billing ceiling per image.

• • The model accepts mixed text and image requests.
• • Each image is capped at 384 billable V4-Flash tokens.
• • The production V4-Flash endpoint remains text-only.
• • Peak uncached input pricing is $0.44 per million tokens and peak output pricing is $1.32 per million tokens.

Source evidence

Confidence 99% / Completeness 92%

What we captured

DeepSeek released V4-Flash-Vision-Exp, an experimental multimodal model available through its API and compatible with OpenAI and Anthropic API formats.

• • The model adds image processing to DeepSeek-V4-Flash.
• • DeepSeek says it nearly matches Opus 4.8 on internal multimodal agent benchmarks.
• • The model supports image inputs and visual agent workflows.
• • Harness version 0.1.1 supports the model out of the box.

Source evidence

Confidence 99% / Completeness 87%

What we captured

Murf AI announced Falcon 2, a proprietary voice foundation model positioned against leading real-time voice models on naturalness, latency, and cost.

• • Murf AI announced Falcon 2 on August 21, 2026.
• • The article reports that Falcon 2 ranked ahead of ElevenLabs Turbo v2.5, ElevenLabs Flash v2.5, and OpenAI Realtime on Artificial Analysis’s Speech Arena.
• • Murf states that Falcon 2 provides sub-100-millisecond time-to-first-audio and costs $0.01 per generated minute.
• • The model supports 150+ voices across 35+ languages and up to 10,000 concurrent calls.

Source evidence

Confidence 99% / Completeness 79%

What we captured

Ornith AI released Ornith-1.5, a three-size open model family designed to generate its own training tasks, scaffolds, and solution rollouts.

• • The release date is August 19, 2026.
• • The family contains 397B MoE, 35B MoE with 3B active parameters per token, and 9B dense variants.
• • Ornith AI says the models generate harder tasks, task-specific scaffolds, and solution rollouts within a reinforcement-learning loop.
• • Weights, serving instructions, and quantized formats were published for developer testing.

Source evidence

Confidence 100% / Completeness 87%

What we captured

OpenAI announced a limited API release of an Ultrafast mode for GPT-5.6 Sol, powered by Cerebras and capable of up to 750 output tokens per second.

• • OpenAI is testing an Ultrafast mode for GPT-5.6 Sol.
• • The mode is available to a small group of customers through the OpenAI API.
• • The mode is powered by Cerebras and can produce up to 750 output tokens per second.

Source evidence

Confidence 94% / Completeness 87%

What we captured

Zhipu announced GLM-5.3 and claimed it outperformed Anthropic's Mythos 5 on a cybersecurity benchmark and narrowed the coding capability gap.

• • Zhipu announced the GLM-5.3 model.
• • Zhipu said GLM-5.3 outperformed Anthropic’s Mythos 5 in a key cybersecurity benchmark.
• • Zhipu said GLM-5.3 closed the gap on coding capabilities.
• • Article date: 2026-08-16.

Source evidence

Confidence 90% / Completeness 70%

What we captured

Z.ai released GLM-5.3, a 1M-context coding and reasoning model, with API access through Together AI serverless and dedicated infrastructure.

• • Together AI lists the endpoint as zai-org/GLM-5.3.
• • Thinking is always enabled, with low, high, and max effort settings.
• • The model is listed with a 1M-token context length.
• • The listed prices are $1.40 per million input tokens and $440 per million output tokens.
• • Weights are described as released following a post-launch safety evaluation, but no current license is specified.

Source evidence

Confidence 99% / Completeness 96%

What we captured

Google launched Gemini 3.7 Flash for coding, agentic, software engineering, web development, and knowledge-work workflows, with availability across its API, enterprise, and consumer products.

• • Google announced the model on August 13, 2026.
• • The introductory price through the end of 2026 is $0.75 per million input tokens and $3.75 per million output tokens.
• • Google reports higher performance than Gemini 3.6 Flash across coding, web development, document reasoning, and business-workflow benchmarks.
• • The model includes updated safeguards for CBRN and cyber-offense misuse domains.

Source evidence

Confidence 100% / Completeness 92%

What we captured

Dealign.ai published Qwen3.8-27B-CRACK-GGUF, a 27B-parameter, safety-modified derivative of Qwen3.8-27B distributed as seven GGUF quantizations and a vision projector for local llama.cpp inference.

• • Repository named Qwen3.8-27B-CRACK-GGUF published and reproduced by dealign.ai
• • Described as a 27B-parameter derivative of Qwen3.8-27B with refusal behavior removed ("abliterated")
• • Repository metadata lists creation date as 2026-08-12
• • Distributed as seven GGUF quantizations (10.5 GB to 29.0 GB) and a 0.9 GB F16 vision projector
• • Recommended local build: 17.0 GB Q4_K_M quantization
• • Packaged for local multimodal inference with llama.cpp and provides OpenAI-compatible local server commands

Source evidence

Confidence 92% / Completeness 91%

What we captured

Nvidia announced Nemotron 3.5 Lightning (30B MoE, 3B active per token), released Aug. 11, with open weights under OpenMDW-1.1 and availability via Nvidia, Hugging Face, and ModelScope; paired with open-source NeMo Switchyard routing library.

• • Model: Nemotron 3.5 Lightning — 30B parameters, mixture-of-experts
• • Active parameters per token: 3B
• • Released Aug. 11 (article text)
• • Context window: up to 1,000,000 tokens
• • License: OpenMDW-1.1, open weights/training data/recipes
• • Availability: Nvidia, Hugging Face, ModelScope; NeMo Switchyard on GitHub

Source evidence

Confidence 95% / Completeness 95%