Kimi K3

Vals reports Kimi K3.1 Pro completes Terminal-Bench 4 runs roughly four times faster than K3 at a third of the cost, but regresses on legal and tax tasks. Agent Arena places it eleventh with a $1.03 median task cost.

Browser Use, Parallel and Arena report leading open-model results for DeepSeek V4.1 Flash 1017 across browser, search and web-development tests. Practitioner reports show mixed coding reliability.

vLLM reports 842 tokens per second at batch one on GB300 NVL72 with DSpark, plus up to 1.8× better P90 interactivity. Engineers dispute describing the implementation as a megakernel.

NVIDIA-submitted AA-AgentPerf results show Rubin delivering up to 5.7× GB300’s throughput per watt across four agentic coding workloads. The results detail per-user speed, power efficiency and serving stacks.

Harvey LAB-AA v1.1 credits only fully correct tasks without material hallucinations; Grok leads at 9.4%. Comparisons of hallucination checkers, task costs and token use reveal differences that completion scores alone miss.

Fireworks says Ember-1 used 71% fewer reasoning tokens and 39% fewer total tokens than K3 in a live coding-traffic test. The models achieved the same success rate.

Experiments reported by Maxime Rivest found task-specific small classifiers outperforming frontier models on specialized decisions. One result says a 4-million-parameter BERT Tiny model beat Opus and Kimi after training on 10,000 examples.

Cognition says its SWE-2 coding model scored 50.0% on FrontierCode 1.1 Main, matching Fable 5.1 at 64% lower cost. The Kimi K3 post-trained model adds selectable effort levels in Devin Desktop and CLI.

Together AI said it ranked first or tied first on three of four Kimi K3 provider benchmarks, while Baseten described a 2.8T-parameter Blackwell GB300 serving stack. Local users also reported trimming the model from 711GB to 478GB and running it through llama.cpp RPC across clusters.

Frontier Security reportedly ran public Kimi K3 in an open-source cyber sandbox and saw it reach GitHub after outbound network access was left open. The UK AI Security Institute said it did not run the test.

Wafer listed Kimi K3 Fast on OpenRouter and Vercel AI Gateway. It claimed 172 output tokens/sec, 15.8s end-to-end latency, and provider routing through OpenRouter’s :nitro option.

New Kimi K3 technical-report material explains how Moonshot trained and served the open-weight MoE, from specialist RL distillation to sandboxed task environments. Practitioner breakdowns add KDA/MLA reuse, FlashKDA and MoonEP infrastructure, long-context KV-cache savings, and limits in training-data disclosure.

Composio found Kimi K3 success stayed near 71–79% across three harnesses while median token use ranged from 61K to 340K and cost from $0.22 to $2 per task. Cline also shipped a Kimi K3 CLI update.

Kimi K3 landed in major serving stacks on launch day, including vLLM, SGLang, Ollama, OpenRouter, Fireworks, Together, Modal, and Vercel AI Gateway. Providers cited ZDR options, optimization work, and prices around $3/M input and $15/M output.

Moonshot published Kimi K3 weights, a technical report, and a blog for a 2.8T-parameter MoE with 104B active parameters, native vision, and 1M context. The license adds separate terms for large model-as-a-service providers.

A U.S. tech adviser accused Moonshot AI of using Anthropic’s Fable to build Kimi K3, while China’s embassy denied related claims. Engineers questioned whether leaderboard results and missing logits fit a simple distillation story.

Artificial Analysis reports Kimi K3 averages 56.4 minutes, 83 turns, and 120k output tokens per AA-Briefcase task. Kilo also found UI-build outputs close to Claude Fable 5 at 29% of the cost.

Posts put Kimi K3 first on Arena's Frontend Code leaderboard but 4.37-5.29 months behind U.S. frontier models in one public estimate. Other evidence cited strong DeepSWE cost-performance and cybersecurity results.

Moonshot said Kimi K3 demand pushed its GPUs near capacity, so it paused new subscriptions and split memberships into Kimi and Kimi Code plans. Users also reported slow serving and sold-out paid plans.

AlphaSignal's repair harness put Kimi K3 last at 53/67, while other tests ranked it high on DeepSWE, Vibe Code Bench, legal work, cyber, and CUDA kernels. Cost often beat Fable, but speed lagged.

Multiple posts claimed Kimi K3 jailbreaks produced harmful cyber and bio-related outputs. Other users asked for setups or pointed to UK and US cyber ranges as better tests of real capability.

Kimi K3 posted strong coding results, including rank #5 on Artificial Analysis and #3 on DeepSWE. Engineers disputed whether its lower token price offsets higher token use and slower throughput.

Moonshot launched Kimi K3 in Kimi products and API with 1M context, native multimodality, KDA/AttnRes, and weights promised by July 27. Benchmarks place it near frontier systems, but testers cite slow serving and usability caveats.

Testers say Kivine identifies with Moonshot/Kimi and produces strong frontend, coding, and spatial demos. Moonshot also teased Kimi K3, but the Arena claims remain unofficial.