Skip to content
Tech wire
CodeSupply 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 Innovationlinuxfoundation.orgAI 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.frThe Open-Source Debt Nobody in Finance Wants to Talk AboutHackerNoonOpen source tech helps farmers put more trust in digital toolsCyprus MailCompare 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)I used a free, open-source AI tool to build a working app – in two daysZDNETSwiss government explores replacing Microsoft 365 with open-source softwareComputerworldThe 10 Coolest Open-Source Software Tools Of 2026 (So Far)crn.com

Llama 4 Maverick

Meta · The open-weights standard-bearer.

Llama 4 anchors the open ecosystem: run it anywhere, fine-tune it freely, pay only for compute — with a 1M-context MoE design.

Live OpenRouter pricing · capability ratings & benchmarks are indicative

Context window

1.0M

tokens

Input · $/1M

$0.188

per million tokens

Output · $/1M

$0.652

per million tokens

Vs. cost floor

2×

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

MMLU-Pro≈77%

solid open-weights showing

LiveCodeBench≈65%

capable coding

MMMU≈74%

multimodal-enabled

Ecosystemleader

most-deployed open family

Where it sits in the field

Output cost per million tokens across all 15 tracked flagships — Llama 4 Maverick 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.

Self-hosted / edge

Sovereign/self-hosted AI

Regulated data that can never leave your infrastructure.

Product AI

Fine-tuned domain models

Own a specialist model instead of renting a generalist.

Agents

Cost-controlled fleets

Fixed-cost inference for always-on agents.

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.