22 results for “MoE”
NVIDIA has released NemotronLabs VoiceChat 11B, an 11-billion-parameter end-to-end speech-to-speech model that performs streaming speech understanding. VoiceChat 11B gives researchers and infrastructure teams a reference implementation.
DeepSeek open-sourced its experimental V4-Flash-Vision-Exp multimodal model on Hugging Face under MIT license on August 31, 2026. The release gives developers local access to the 305B-parameter vision-language model that previously ran only on DeepSeek's API platform.
Mistral AI has released Shieldstral, a 3-billion-parameter open-weight content moderation model that the company says matches the safety performance. Content moderation has traditionally forced a trade-off between coverage and cost.
A JPMorgan note relayed by Reuters and CGTN forecasts a Zhipu AI model exceeding 1 trillion parameters for August 2026. Zhipu has not confirmed the model, its specs, or the date — leaving developers to separate analyst signals from vendor commitments.
Z.ai released GLM-5.2, a 753B open-weight model with MIT licensing that matches Anthropic's Opus 4.8 on agentic benchmarks at roughly one-fifth the token cost. The 1M-context model is already running in Cursor and Claude Code, shifting enterprise leverage toward self-hosted inference.
NVIDIA's Groq 3 LPX accelerator hit 3,431 output tokens per second on the Artificial Analysis 100K context benchmark with Gemma 4 31B. The result enables responsive multi-agent systems that can process full session histories at long context lengths.
DeepSeek, Qwen, and GLM-class APIs often price an order of magnitude below Claude Opus, GPT-5.6 Sol, and peers. The useful question is which tier matches the job you intend — and how close the outcome gets without paying for unused ceiling.
Anonymous model Ox Alpha launched on OpenRouter August 20 with a 1M-token context window and free preview access. Independent testing on a 10-task DeepSWE subset showed an 80% pass rate, beating Claude Fable 5, GLM-5.3, GPT-5.6 Sol, and Grok 4.6 on the same tasks.
Zhipu AI compressed its model releases to two-month intervals across early 2026, culminating in GLM-5.2's June launch with a 1-million-token context window. The cadence signals a production-line strategy that holds parameter count steady while iterating context, tooling, and pricing to undercut U.S. rivals on agentic coding workloads.
Moonshot AI has launched Kimi K3, a 2.8-trillion-parameter model with native multimodal capabilities and a 1-million-token context window. Open-model releases have mostly clustered well below the trillion-parameter class.