Together AI – a provider of AI inference and tooling infrastructure – saw an active week of product and partnership updates, underscoring a strategic focus on cost-efficient, voice-centric, and agentic workloads. This recap reviews the key developments and their implications for the company’s competitive positioning.
Meet Samuel – Your Personal Investing Prophet
- Start a conversation with TipRanks’ trusted, data-backed investment intelligence
- Ask Samuel about stocks, your portfolio, or the market and get instant, personalized insights in seconds
Together AI announced a partnership with Pearl Research Labs to launch the Gemma‑4‑31B‑it‑Pearl model as its first Pearl‑powered endpoint. The model is offered at more than 25% discounted pricing, leveraging Pearl Network’s Proof of Useful Work protocol and ¶PRL token economics to potentially reduce compute costs over time.
Gemma‑4‑31B‑it‑Pearl supports a 256K context window, configurable “thinking” for step‑by‑step reasoning, function calling, JSON mode, and multimodal capabilities on some variants. These features are aimed at AI‑native developers and enterprises that require long‑context, tool‑integrated and multimodal workloads at lower effective prices.
The company also deepened its text‑to‑speech offering by hosting Rime Mist v3 models on dedicated infrastructure. Two endpoints support English and multilingual voice agents, with deterministic pronunciation, domain‑specific pronunciation controls, SSML‑based pacing, and concurrency for high‑throughput use cases.
These TTS capabilities target enterprise voice agents in sectors such as finance, healthcare, and technical support, where precision and reliability are critical. Multilingual support in English, Spanish, French, and German broadens Together AI’s appeal to global customers and may increase platform stickiness.
Complementing its TTS expansion, Together AI introduced a “voice finder” discovery tool aggregating more than 600 voices from providers including MiniMax, Cartesia, Deepgram, and Rime. Developers can search by prompt or audio sample and filter by attributes like pitch, accent, language, age, emotion, and speaking style.
By acting as an aggregation and orchestration layer across multiple TTS vendors, Together AI is moving closer to end‑application design. This tooling could drive higher engagement, deepen third‑party partnerships, and support incremental usage‑based revenue in AI agent and voice interface markets.
On the speech‑to‑text side, Together AI highlighted that its hosted models rank first and second for transcription speed on the Artificial Analysis leaderboard. NVIDIA Parakeet TDT 0.6B V3 reportedly processes 303 seconds of audio per second of compute at about $1.50 per 1,000 minutes, with an AA‑WER of 4.6% across real‑world datasets.
This combination of speed, cost, and accuracy positions Together AI’s STT stack as core infrastructure for real‑time voice applications. Performance leadership may enhance its appeal to latency‑sensitive AI‑native products and support usage growth in emerging voice agent segments.
The company also showcased technical advances in serving the DeepSeek V4 Pro model on its serverless platform. A technical deep dive detailed KV cache compression techniques, multiple cache layouts, prefix caching, and workload‑specific endpoint tuning aimed at improving efficiency for long‑context and coding workloads.
These optimizations point to continued investment in scalable, high‑performance inference infrastructure. Improved unit economics and guidance on benchmarking may help attract enterprise customers seeking cost‑effective hosting of state‑of‑the‑art third‑party models.
Finally, Together AI highlighted its role powering browser‑based AI agents developed by Yutori, including continuous web monitoring and AI “chief of staff” workloads. Reported results include roughly 2x faster per‑step inference than leading frontier models, 4–5x lower inference costs, 99.9% uptime, and elastic scaling without contract renegotiation.
If representative of broader deployments, these metrics reinforce Together AI’s positioning as a reliable, cost‑efficient platform for high‑volume, agentic workloads. Overall, the week’s updates emphasize differentiated pricing, advanced voice capabilities, and technical depth in model serving, collectively strengthening the company’s infrastructure value proposition and potential for sustained usage‑based growth.