Skip to content
Tech wire
Businesses Embrace Open-Weight AI Amid Heavy Tech CostsPYMNTS.comCodeSupply Announces R and D Grants for Open-Source Software SecurityfundsforNGOsOpenAI bots meddled with US government agencies, including SEC and CensusBBCOCUDU Ecosystem Foundation Welcomes GNU Radio to Accelerate Open Source Wireless InnovationLinux FoundationAI Is reshaping open source software and straining the systems that sustain itEurekAlert! Science News ReleasesOpen-source coding platform makes radar technology accessible to scientists worldwideUniversity of Colorado BoulderArthur Mensch, CEO of French start-up Mistral AI: 'AI is software. It can be controlled'Le Monde.frCompare Top 21 Manufacturing AI Solutions & SoftwareAIMultipleAI agents being tested by OpenAI involved in cyber-attack on another service, say researchersThe GuardianNIST Updates 5G Open-Source Testbed ToolsNational Institute of Standards and Technology (.gov)Open source tech helps farmers put more trust in digital toolsCyprus MailI used a free, open-source AI tool to build a working app – in two daysZDNETThe 10 Coolest Open-Source Software Tools Of 2026 (So Far)crn.comSwiss government explores replacing Microsoft 365 with open-source softwareComputerworld

Kimi K3

Moonshot · Open agentic muscle.

Kimi's K-line is the open ecosystem's agentic specialist — trained hard on tool use and long-horizon tasks, at value pricing.

Live OpenRouter pricing · capability ratings & benchmarks are indicative

Context window

1.0M

tokens

Input · $/1M

$3.00

per million tokens

Output · $/1M

$15.00

per million tokens

Vs. cost floor

52×

output cost vs DeepSeek V4.1 Flash

Capability profile

Editorial ratings across six axes (0–100), compiled from publisher reports and public evals — indicative, not measured by TrendNexus.

Benchmark highs

SWE-bench Verified≈68%

strong open-agentic result

Tau-bench (tools)≈75%

tool-use reliability

AIME-class math≈88%

solid math

MMLU-Pro≈81%

broad knowledge

Where it sits in the field

Output cost per million tokens across all 15 tracked flagships — Kimi K3 highlighted.

Claude Fable 5.1$50.00GPT-5.5$30.00Claude Opus 5.5$20.00Kimi K3$15.00GPT-5.6 Terra Pro$12.00Gemini 3.1 Pro Preview$12.00GPT-6 Sol Pro$10.00Claude Sonnet 5$10.00GLM 5.3 Prime$8.80Grok 4.7$4.80Qwen3 Max Thinking$3.90Gemini 3.8 Flash$3.75DeepSeek V4 Pro 0423$0.696Llama 4 Maverick$0.652DeepSeek V4.1 Flash$0.29

What it's suited to build

Product classes where this model's profile wins — mockups are illustrative, not model output.

Agents

Open-stack agent systems

Agentic backbones without closed-model dependency.

Coding

Automation engineering

Scripted operations and integration glue at low cost.

Self-hosted / edge

Hybrid deployments

API or self-hosted as economics dictate.

Benchmark figures and capability ratings are indicative — compiled from publisher reports and public leaderboards, refreshed editorially, and not independently measured by TrendNexus. Verify against your own evals before procurement decisions. Pricing updates with the daily build.