Together AI continued to build out its AI Native Cloud this week, adding two production-ready third-party models aimed at advanced coding and agentic workloads. The company emphasized reliability with 99.9% SLAs and serverless deployment options designed for AI-native developers and enterprises.
Claim 55% Off TipRanks
- Unlock hedge fund-level data and powerful investing tools for smarter, sharper decisions
- Discover top-performing stock ideas and upgrade to a portfolio of market leaders with Smart Investor Picks
Together AI introduced DeepSeek V4 Pro, a long-context reasoning and coding model optimized for complex software tasks and agentic workflows. The model supports up to a 512K token context window and offers three reasoning modes that allow users to balance speed against depth of analysis.
Benchmark results highlighted in the company’s materials include 93.5% on LiveCodeBench, a 3206 Codeforces rating, and 80.6% on SWE-Bench Verified. Together AI also cited efficiency gains over the prior V3.2 version, including reduced FLOPs and KV cache usage for long-context inference, which could help customers manage costs.
A related DeepSeek V4 Flash variant is described as coming soon, signaling an expanding roadmap around this model family. The broader DeepSeek offering is positioned to capture workloads that require long-horizon coding, high-context reasoning, and production-grade reliability on Together AI’s infrastructure.
In parallel, Together AI rolled out access to Moonshot AI’s Kimi K2.6, a multimodal and agentic model designed for autonomous workflows. The model can coordinate up to 300 sub-agents executing as many as 4,000 steps, targeting complex orchestration scenarios and long-horizon coding tasks.
Kimi K2.6 supports text, image, and video understanding and is offered via both serverless and dedicated deployments on the AI Native Cloud. Performance benchmarks, including SWE-Bench Verified, LiveCodeBench v6, and MMMU-Pro, are used to position the model for demanding enterprise use cases.
These additions underscore Together AI’s strategy of curating advanced third-party foundation models rather than focusing solely on in-house model development. By emphasizing reliability, efficiency, and support for agentic and multimodal workflows, the company aims to deepen engagement with AI-native developers.
From a financial perspective, the expanded model catalog could increase usage-based revenue and improve customer retention if adoption scales. Overall, the week marked a strengthening of Together AI’s competitive stance in the AI infrastructure and model-hosting market, with a clear focus on high-value, production-ready workloads.