Baseten上线Kimi K3模型API，提供首日访问

Kimi K3 launched with day-0 partners including Nebius, Baseten, Fireworks AI, DigitalOcean Serverless Inference, and Together AI, per Moonshot's launch-day announcements. Kimi K3 launched with Max reasoning effort only, then added Standard and High effort levels within roughly a week, per TestingCatalog. Kimi K3 scores 57 on the Artificial Analysis Intelligence Index, ranking fourth behind Claude Opus 5, Claude Fable 5, and GPT-5.6 Sol. Kimi K3 costs roughly $0.94 per Intelligence Index task versus GPT-5.6 Sol at $1.04 despite lower per-token pricing, driven by verbose output, per Artificial Analysis. [...] Kimi K3 is a 2.8T-parameter MoE with 16 of 896 experts active per token and ~2.5× better scaling efficiency than Kimi K2, per Moonshot's official blog. New architecture components: Kimi Delta Attention (KDA), Attention Residuals (AttnRes), and a Stable LatentMoE framework. Open weights shipped July 27, 2026 to Hugging Face — exactly on the date Moonshot committed to at launch.

Kimi K3 is Kimi’s most capable flagship model to date, with 2.8 trillion parameters. It is built on Kimi Delta Attention (KDA), a hybrid linear attention mechanism, and Attention Residuals, with native visual understanding and a 1M-token context window. It is the world’s first open-source model in the 3-trillion-parameter class, designed for frontier intelligence scenarios including long-horizon coding, knowledge work, and reasoning.For complete benchmarks and case studies, see the technical blog. Kimi is currently working closely with inference partners and open-source maintainers to align technical details and ensure the model launches reliably across the ecosystem. The full model weights will be released by July 27, 2026. More details on architecture, training, and evaluation will be [...] A 3-trillion-scale open-source model

``` [...] ``` [...] --speculative-config '{"model":"Inferact/Kimi-K3-DSpark","method":"dspark","num_speculative_tokens":7,"attention_backend":"FLASHINFER_MLA","draft_sample_method":"probabilistic","rejection_sample_method":"block"}' # Batch size = 1, 8K/1K random (no speculative decoding)# Batch size = 1, 8K/1K random (no speculative decoding)vllm-bench \vllm-bench \ --backend openai \ --backend openai \ --base-url " \ --base-url " \ --model moonshotai/Kimi-K3 \ --model moonshotai/Kimi-K3 \ --dataset-name random \ --dataset-name random \ --random-input-len 8192 \ --random-input-len 8192 \ --random-output-len 1024 \ --random-output-len 1024 \ --random-range-ratio 0.8 \ --random-range-ratio 0.8 \ --prompt-token-ids \ --prompt-token-ids \ --ignore-eos \ --ignore-eos \ --sweep-max-concurrency 1 \

Kimi K3 is available today on Baseten Model APIs. Welcome to the new frontier in open weight intelligence. Subscribe to our newsletter Stay up to date on model performance, inference infrastructure, and more. ## Explore Baseten today Product Dedicated Inference Model APIs Training Model Labs Baseten Inference Platform Model Runtimes Infrastructure Multi-cloud Deployment options Cloud Self-hosted Hybrid Embedded engineering Forward deployed engineers Modalities Transcription Image Generation Text-to-speech Large language models Compound AI Embeddings Industries Enterprise Healthcare Developer Model library Documentation Changelog Resources Research Customers Blog Guides Events Partners Savings calculator Trust Center About us Startups Careers [...] ## Build with Kimi K3 on Baseten We’re excited to offer day-0 access via Model APIs, and look forward to continuing to optimize our implementation of this model to achieve the highest standards in

## Fireworks AI# Fireworks exposes K3 as `accounts/fireworks/models/kimi-k3`. It supports serverless inference, on-demand dedicated GPUs, image input, function calling, and fine-tuning. The differentiator is operational control. Fireworks says its K3 serverless endpoint is US hosted with zero data retention and offers a path from shared inference into dedicated capacity and tuning. ## Baseten# Baseten's K3 Model API supports image input and the full 1M-token context through an OpenAI-compatible endpoint. Its day-0 article is the clearest infrastructure explanation in this provider set, including validation with the Kimi Vendor Verifier, vLLM, and SGLang. [...] Moonshot Kimi K3 technical blog - fetched July 27, 2026 Kimi K3 API quickstart - fetched July 27, 2026 Kimi K3 Hugging Face model card - fetched July 27, 2026 Together AI Kimi K3 model page - fetched July 27, 2026 Fireworks Kimi K3 model page - fetched July 27, 2026 Fireworks Kimi K3 launch article - fetched July 27, 202

Kimi.ai @Kimi\_Moonshot Jul 27 Kimi K3 is now live on @togethercompute! Happy to have Together AI as our day0 launch partner, giving developers immediate access to K3 through high-throughput inference optimized for coding agents and production workloads. A huge thank-you to the Together AI team for user avatar Together AI @togethercompute Jul 27 Kimi K3 is now live on Together AI. We’re proud to be a Day 0 launch partner for @Kimi\_Moonshot’s open frontier model, built for long-running agentic workflows across code, tools, vision, and research. 112K user avatar Kimi.ai @Kimi\_Moonshot Jul 27 Kimi K3 is now available on @digitalocean 's Serverless Inference! Developers can start building with our most capable model in minutes.