CheckCom AI Model Release Radar

New AI models & LLM versions

CheckCom evaluates ongoing AI news, the model catalog and pricing signals. New models, versions, previews, API access, open-weight and local/quantized candidates become visible early.

Current overview

2detected today
650release candidates
12providers / authors
2new in catalog
328pricing relevant
251local / quantized

Stand: 22.08.2026 · Datenbasis: 813769 News-Einträge und 2 Modellkatalog-Einträge.

Enterprise-relevant model and version signals

new model or versionstrongpricing

Gemini 3.7

Google / Gemini · 22.08.2026

On August 13, 2026, Google introduced Gemini 3.7 Flash, a cost-effective AI model for coding and automated workflows. Additionally, the company integrated Antigravity, its agentic development platform, into Gemini Enterprise subscriptions.

New model/version signal: relevant for roadmap, vendor shortlist and migration planning. Potentially relevant for coding or agent workflows. Pricing or billing signal detected; check price radar and primary source.

ecosistemastartup.com

new model or versionstrongpricing

Gemini 3.1

OpenAI · 22.08.2026

RayNeo has launched new AI glasses that skip the camera and speakers. The RayNeo iO Glasses overlay text in the user's field of view and use AI models such as RayNeo AI and Gemini 3.1 Flash Lite. A subscription provides access to additional models like ChatGPT 5.1.

New model/version signal: relevant for roadmap, vendor shortlist and migration planning. Potentially relevant for coding or agent workflows. Pricing or billing signal detected; check price radar and primary source. EU/DACH or availability signal detected; check region, data residency and contract route.

the-decoder.com

preview or experimentalstrongpricing

DeepSeek V4-Flash-Vision-Exp

Anthropic · 21.08.2026

On August 21, 2026, DeepSeek introduced an experimental multimodal AI model, V4-Flash-Vision-Exp, offering performance comparable to Anthropic's Opus-4.8 at just 1/35 the cost. The model can process images, screenshots, and graphics, making it suitable for automated workflows. It presents a cost-effective alternative for startups integrating AI into their products.

Preview or experimental status: production readiness and stability should be checked separately. Multimodal use may create additional pricing, privacy and data-format questions. Pricing or billing signal detected; check price radar and primary source.

ecosistemastartup.com

preview or experimentalstrong

Deepseek

Anthropic · 21.08.2026

DeepSeek has released V4-Flash-Vision-Exp, an experimental multimodal model that adds image understanding to V4-Flash's text capabilities. On the company's own benchmarks, it approaches Opus 4.8 and sometimes outperforms it.

Preview or experimental status: production readiness and stability should be checked separately. Potentially relevant for coding or agent workflows. Multimodal use may create additional pricing, privacy and data-format questions.

the-decoder.com

availability or api updatestrong

Gemini Omni

Google / Gemini · 20.08.2026

Adobe is making three AI-powered audio tools widely available in its Firefly platform. Generate Music, Generate Speech, and Generate Sound Effects create royalty-free music, voiceovers, and sound effects for video projects. Google's Gemini Omni Flash has also been added to the platform.

Potentially relevant for coding or agent workflows. Multimodal use may create additional pricing, privacy and data-format questions. EU/DACH or availability signal detected; check region, data residency and contract route.

the-decoder.com

new model or versionstrong

DeepSeek Harness Launches

Anthropic · 20.08.2026

DeepSeek has launched Harness, a new AI model, and compared it with Grok Build. The question of whether these models could rival Claude's code model is being discussed.

New model/version signal: relevant for roadmap, vendor shortlist and migration planning. Potentially relevant for coding or agent workflows. EU/DACH or availability signal detected; check region, data residency and contract route.

pub.towardsai.net

local or quantized updatestrongpricinglocal/quant

Kimi K3 and GLM-5.3 are now

Moonshot / Kimi · 20.08.2026

China's latest AI models, such as Kimi K3 and GLM-5.3, are now close to the best US models. Western labs accuse China of distillation, but regardless of guilt, the conclusion remains: a model lead is hard to defend.

Local or quantized use is mentioned: check hardware requirements, license and measurement status. Potentially relevant for coding or agent workflows. Pricing or billing signal detected; check price radar and primary source. EU/DACH or availability signal detected; check region, data residency and contract route.

the-decoder.com

new model or versionstrongpricinglocal/quant

GPT-5.6

OpenAI · 20.08.2026

Amazon Web Services (AWS) has made OpenAI's GPT-5.6 Terra and Luna models generally available in India via Amazon Bedrock. The models are processed locally and offer up to 80% cost reductions. They support various applications, including code generation and automated reporting.

New model/version signal: relevant for roadmap, vendor shortlist and migration planning. Potentially relevant for coding or agent workflows. Pricing or billing signal detected; check price radar and primary source.

newsbytesapp.com

new model or versionstrongpricing

Grok 4.6 Grok 4.6

xAI / Grok · 20.08.2026

Grok 4.6 from xAI introduces major upgrades for agentic workflows, software engineering, and interactive web application generation. It maintains context over extended periods and self-verifies its work before proceeding.

New model/version signal: relevant for roadmap, vendor shortlist and migration planning. Potentially relevant for coding or agent workflows. Pricing or billing signal detected; check price radar and primary source.

producthunt.com

new model or versionstrong

Gemini Launches

Google / Gemini · 19.08.2026

Google has introduced a new Student Hub in Gemini featuring AI-powered study tools such as flashcards, practice quizzes, and smart calendar integration. The tool is designed for the education market and competes with platforms like Quizlet and Notion.

New model/version signal: relevant for roadmap, vendor shortlist and migration planning. Multimodal use may create additional pricing, privacy and data-format questions. EU/DACH or availability signal detected; check region, data residency and contract route.

techbuzz.ai

availability or api updatestrongpricinglocal/quant

Gemma 4

Google / Gemini · 19.08.2026

A technique called Multi Token Prediction can significantly increase the generation rate of local LLMs. Tests showed up to 250 percent more tokens per second. The feature is not available for every model, but works with models like Gemma 4, Qwen3.5, and Qwen3.6.

Local or quantized use is mentioned: check hardware requirements, license and measurement status. Pricing or billing signal detected; check price radar and primary source. EU/DACH or availability signal detected; check region, data residency and contract route.

heise.de

new model or versionstrong

Gemini

OpenAI · 19.08.2026

Google has introduced new AI-powered study tools in Search and Gemini, including interactive visuals, 3D simulations, and personalized quizzes. The move aims to position Gemini as a go-to AI assistant for students and to strengthen competition with companies like OpenAI.

New model/version signal: relevant for roadmap, vendor shortlist and migration planning. Multimodal use may create additional pricing, privacy and data-format questions. EU/DACH or availability signal detected; check region, data residency and contract route.

techcrunch.com

New in the CheckCom model catalog

DateModel / signalProviderAccessLocal suitabilityEnterprise note
28.05.2026Qwen2.5-Coder 7B Instruct
alibaba-qwen-qwen2-5-coder-7b-instruct
Alibaba / Qwenopen weightlocal edgelocal candidateyesNew in the model catalog. Enterprises should check source, license, price route, local suitability and model quality before production use.
28.05.2026Meta Llama 3.1 8B Instruct
meta-llama-3-1-8b-instruct
Meta / Llamaopen weightlocal edgelocal candidateyesNew in the model catalog. Enterprises should check source, license, price route, local suitability and model quality before production use.

How CheckCom classifies these signals

The Release Radar combines news signals, the model catalog, newly detected catalog entries and pricing-radar signals. An entry does not automatically mean that a model is production-ready, legally cleared or fully price-confirmed. CheckCom therefore marks source, signal type, pricing relevance, local/quantized hints and catalog matching.

  • Confirmed: strong source or catalog match available.
  • Candidate: detected from news; primary source or catalog match should still be checked.
  • Pricing-relevant: the item contains price, API, token, availability or billing signals.
  • Local / quantized: hints for open weights, GGUF, Ollama, quantization or edge hardware.

Additional detail reports

For customers, these data can be turned into detailed provider, model, pricing and migration reports. The public page shows a condensed market view; internal raw data and full candidate lists are not offered as public downloads.

Request analysis

Understanding AI models

New LLM versions, quantization, local AI, Raspberry Pi, workstation hardware and AI costs explained — with links to the key CheckCom radars.

Open guide