Artificial Analysis says Google has released Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, Google DeepMind text-to-speech models with preset voices and voice replication. On its tests, Gemini 3.8 Flash TTS scores 89.5% in pronunciation robustness and an Elo of 1,263 on the Provider Voice Arena.
Google reports a 71.4 overall score and a 60.8 accent-modeling score for Gemini 3.8 Flash TTS on Hume AI Voice Design. It also places Flash TTS first and Gemini 3.8 Flash-Lite TTS second on Hume AI Overall Quality.
DAIR.AI reports that a Stanford and Together AI paper found o3-mini, Claude Sonnet 4, and DeepSeek-V3 averaged 66.7% across five math and physics benchmarks as a self-organizing team, above 48.8% for the strongest member alone and 59.0% for a perfect router.
Artificial Analysis finds no single open-weight text-to-speech model ahead on all nine new language leaderboards. Leadership is split among Higgs Audio V3 TTS, OpenAudio S1 Mini, Voxtral TTS and Breeze TTS 2.
Artificial Analysis reports a transcription speed factor of 88x real time for StepAudio 3 ASR, behind three faster systems, near ElevenLabs Scribe v2 and Gemini 3.5 Transcribe, and ahead of Fun-Realtime-ASR-preview.
OpenAI reports that GPT-6 Luna scores 66.6% on DeepSWE v1.1, comparable to Claude Opus 5 and Fable 5 at medium effort for 93-96% less per task, and matches GPT-5.6 Sol on factuality at higher effort for about 1/100th the cost.
On September 22, 2026, Scale AI released SWE-Bench Pro V2 as a refreshed public split of SWE-Bench Pro.
elvis (@omarsar0) shared a paper on EvoOntology, a self-evolving ontology layer for data agents served as an MCP server. The post says DDR-Bench accuracy rises 17.8 points on average across six backbones, including 26.7 points for GPT-5.5, and BIRD execution accuracy rises 7.4 points.