The contest tested approaches to transcribing Lingala and Shona with open speech data. The next question is whether those results can translate into useful voice services.

2 min

Sep 22, 2026Models

Research Models

Researchers Release C2C to Let AI Models Share Internal Memory Instead of Text

The benchmark gains are promising for multi-model systems, but the technique requires direct access to model internals and has not been tested on production agent workloads.

3 min

Sep 22, 2026Models

Launch Models

OpenAI Releases GPT-6 Sol and Luna With API Prices Half Their Predecessors’

The release creates two cheaper options below GPT-6 Astra, but OpenAI’s promised accuracy gains remain based on its own tests.

3 min

Sep 22, 2026Models

Launch Models

GitHub Adds GPT-6 Sol and Luna to Copilot, Giving Coders More Model Choices

The gradual rollout adds a balanced option for multistep coding and a lower-cost option for smaller work, while organizations retain control over access.

3 min

Sep 22, 2026Models

Launch Models

Anthropic Releases Claude Opus 5.5, Cutting Prices From Opus 5

The new flagship model is Anthropic’s bet that agentic work will be won as much by lower serving costs and faster output as by benchmark gains.

3 min

Sep 22, 2026Models

Launch Models

TBC Brings Rat-Brain Video AI to Select AWS Customers

The limited preview gives TBC a new route into AWS for software meant to make video generation faster and cheaper. Whether those gains persist in longer videos remains unresolved.

2 min

Sep 22, 2026Models

Launch Models

Austrian Academy Develops Apollo to Reconstruct Damaged Ancient Greek Texts

The planned free academic chatbot is designed to speed a painstaking scholarly task without turning probability-based AI output into the historical record.

3 min

Sep 22, 2026Models

Launch Models

China Telecom AI Releases 29B Agent Model for Single-GPU Deployment

Xing4.0-29B-A4B uses a sparse design that activates 4 billion parameters at a time. China Telecom AI says that lets developers run long-context, tool-using workflows locally with about 15 GB of GPU memory.

2 min

Sep 21, 2026Models

Launch Models

Tencent Releases Hy Image 3.5 Preview With Multi-Round Editing and 4K Output

The new image model reaches Tencent’s chatbot, creative tools and cloud API. Its claimed competitive results come from internal designer testing, leaving independent quality comparisons unresolved.

2 min

Sep 21, 2026Models

Launch Models

Xiaomi Releases MiMo-V2.6 Models as Its Pro Version Tops an Open-Weight Ranking

The new family pairs a high-ranking flagship with a cheaper Flash model and an accelerated serving variant, but Xiaomi’s benchmark results still leave real-world agent reliability to be tested.

4 min

Sep 21, 2026Models

Launch Models

StepFun Releases Step 5 Preview With 1M Context for Agent Work

The new model separates its 600 billion total parameters from the 27 billion it activates per token, but developers can use it only through StepFun’s hosted service for now.

3 min

Sep 21, 2026Models

Launch Models

xAI Releases Grok 4.7 With Longer Reasoning at Grok 4.6 Prices

The new model is available through xAI’s API, Cursor and Grok Build, but its performance case depends on company results that do not fully line up with a published independent summary.

4 min

Sep 21, 2026Models

Open release Models

Microsoft Research Open-Sources RetroChimera for Chemical Synthesis Planning

The newly published system ranks free-form molecular predictions against template-guided ones, a design intended to handle both familiar and rare reactions.

2 min

Sep 21, 2026Models

Open release Models

Yandex Releases Alice AI Foundation, an 80B Model for Commercial Use

The downloadable base model uses only 3 billion parameters per token, comes with Russian-language factual benchmarks, and sits ahead of Yandex’s planned reasoning and agent work for Alice AI.

2 min

Sep 21, 2026Models

Launch Models

Aikido Releases Altar, a Local Cybersecurity AI Model Built From GLM-5.3

The open-weight release is designed to keep sensitive code within customer infrastructure, but its claimed security performance comes from a narrow internal benchmark.

2 min

September 2026

1 story

Sep 20, 2026Models

Open release Models

Alibaba Releases Qwen-Image-2.1 With Transparent Image Editing

The 7-billion-parameter model gives developers a single open-weights package for generation, editing, transparency, and reference-guided work. But businesses need a separate license before using it commercially.

2 min

Load more stories18 more next