IBM Granite Speech 5.0 2B, an Apache 2.0 ASR model, ranks #2 on Open ASR Leaderboard at 5.68%. IBM speech recognition model ...

Recursive Flint is a 40B MoE local coding model with 2.5B active parameters, scoring 74.2% on SWE-bench Verified.

Vercel open-sourced durable-harness, an Apache-2.0 package that makes Pi, OpenCode, Claude Code and Codex agent sessions ...

Perplexity’s JOLT trains one model as teacher and student, matching GRPO’s MATH accuracy with 6.5x fewer training tokens.

Google open-sourced TorchTPU, a native PyTorch backend for Cloud TPUs that reaches 52% MFU training Qwen 3 8B.

DeepSeek open-sourced DeepInfer, its production inference framework for DeepSeek-V4.1-Flash, reporting over 25,000 decode ...

Exa 3.0 search API launches with Exa Lite at $4 per 1,000 requests and leads 8 self-reported benchmarks (AI search API ...

Cloudflare’s Clairic extraction model returns typed values in one pass, cutting a claims pipeline’s run time by 69%.

Google team have released llm-d 1.0, the first stable version of its distributed inference stack. This open-source, ...

Mistral Voxtral 3 adds speech-to-text, text-to-speech and a voice agent SDK, with transcription from $0.0035 per audio minute ...

Mistral AI released Voxtral 3, a family of 3 transcription models, 1 text-to-speech model and a LiveKit-based voice agent SDK. Voxtral Small Transcribe 3 scores 2.1% AA-WER, ranking #3 on Artificial ...

Google and Meta open-sourced TorchTPU, a native PyTorch backend for Cloud TPUs. It runs eager by default, compiles through torch.compile and XLA, and replaces PyTorch/XLA for new workloads.