Meta just turned your usage data into a bargaining chip. For its new Muse Spark model, built to power coding and other agents, the company is offering a discount that averages around 95% to users who agree to “contribute” their prompts and model outputs to training future versions. That’s the news, according to TechCrunch AI, and it flips the usual privacy pitch on its head. Instead of letting you opt out of data sharing for free, Meta is paying you to opt in.
The math is blunt.
💰 What the pricing looks like
- Standard rate: 1 million input tokens cost $1.25. Under the contributor tier, the same million costs 10 cents.
- Output tokens: $4.25 per million at standard pricing, versus just 20 cents for contributors.
- Meta’s own pricing guide says the contributor tier “lowers the barrier to entry for prototyping, testing integrations, and scaling experiments where training on your data is acceptable.”
So you’re looking at roughly a 90-95% cut in exchange for handing over how you actually use the model. When TechCrunch AI asked Meta about the new pricing, the company didn’t respond.
🧭 Why Meta is doing this
Meta has struggled to get quality training data for agentic tools. Earlier this year it launched an effort to track its own employees’ computer usage, which drew heavy internal criticism and got paused in June. This pricing model is a cleaner path to the same goal: buy the data instead of harvesting it internally.
And that data really matters. Mario Zechner, the developer behind the open source harness Pi, told TechCrunch AI that the jump in coding agent capability between April and October 2025 came because “Claude Code, by default, would store all your coding agent sessions and use them for reinforcement learning training.” Real usage traces are the fuel. The problem is that most professional workflows outside software engineering leave few digital traces, which makes agents hard to evaluate and improve.
🏢 The enterprise wrinkle
What stands out here is who Meta is really courting. Big companies guard their data. Princeton computer science professor Arvind Narayanan pointed out that large firms “stick with token-billed Enterprise plans even though the subscription-based consumer plans like Claude Max and ChatGPT Pro are discounted by 10x-20x or even more,” mostly because those enterprise plans keep their data out of training.
Meta’s move puts a number on that trade-off. By offering explicit compensation, it’s nudging companies to sort out which data is genuinely proprietary and which they’d happily share for a 95% discount. Narayanan suggested that pressure could push firms to get more precise about what’s actually sensitive.
📉 The bigger price war
This also lands in the middle of a pricing brawl between the frontier labs. Anthropic’s newest Fable and Mythos models arrived with lower costs for processing cached tokens, and OpenAI cut prices sharply at the end of July. Meta’s contributor tier is a different weapon in the same fight: it competes on price while solving a data problem at the same time.
What to watch next
- Expect other labs to test similar “pay-with-your-data” tiers if Meta sees uptake. It’s an elegant way to lower headline prices without eating the full cost.
- If you’re a developer or a small team, the discount is real money, but read the terms. Your prompts and outputs become training material.
- If you’re at a larger company, this is a prompt to actually audit your data. Some of it is worth guarding. Some of it is just a discount you’re not claiming.
The interesting shift here is philosophical. Data sharing used to be the default you quietly accepted. Meta is treating it as a market, with a posted price. Watch whether that framing spreads, because once one lab puts a dollar figure on your usage, the rest tend to follow. Full details are at TechCrunch AI.