Tether’s 800M AI Agent Beats Models 2.5x Its Size on Everyday Phone Tasks
- Published
- Updated
Tether has introduced BotPsy-0.8B, an 800-million-parameter open-weight model for local AI agents that can handle multi-step phone and tool tasks. Across benchmark coverage from several outlets, the model is described as outperforming larger systems in Tether’s tests, including claims that it beat a model 2.5 times its size and roughly doubled its base architecture’s tool-use scores.
Key claims & sources
What it means
Tether built its name on USDT, but its AI Research arm has spent the past year pushing a parallel bet: that on-device, open-weight models can replace cloud-dependent AI agents. BotPsy-0.8B fits that strategy, an 800-million-parameter release aimed at phones rather than data centers, pitched against rivals such as Qwen3.5-2B-Think that run on far more parameters.
Developers building phone assistants can now test a model small enough to run locally, skipping the API calls and licensing fees tied to larger cloud systems. Device makers gain a path to multi-step tool tasks, like booking or form-filling, without the compute budget those tasks usually demand. App teams chasing lower hardware requirements have a concrete alternative to weigh against bigger generalist models. (per cryptobriefing.com)
A stablecoin issuer releasing a competitive open-weight model says something about where Tether wants to sit in AI infrastructure, not just payments: closer to Big Tech's cloud AI than to its own core business. Beating a model 2.5 times its size on vendor-run benchmarks is a strong claim, and claims like that earn scrutiny until independent labs reproduce them outside Tether's own test suite.
The test that matters now is replication: whether outside researchers can reproduce BotPsy's wins against Qwen3.5-2B-Think on the same six agentic benchmarks, or whether the doubled tool-use scores hold only inside Tether's own four-benchmark set. Open weights make that check possible immediately; what happens when independent labs run it will decide if the result is a genuine shift or a vendor benchmark.
Market reaction i
Tether barely moved in the 4 hours after the first headline, after a flat 2 hours before.
48 prices, one every 30 minutes
Coverage timeline
Every piece in order, with the time its news medium declares — if it is not believable we use ours and say so — and what it added to what was already known at that point. Arriving first is not the same as adding something.
Has this happened before?
| When | What happened | Figure |
|---|---|---|
| 8 days ago | Justin Sun launches tokenized US stock trading on HTX with 20 assets event | 1 actors in common · 2 news media |
| 9 days ago | Kinetiq launches Elysium Layer 2 to expand Hyperliquid beyond perps event | 1 actors in common · 4 news media |
| 15 days ago | Coinhub Funded Goes Live: Crypto Prop Firm Offers Funded Trading Accounts Up to $200K & 30-Minute Tournaments event | 1 actors in common · 3 news media |
| 20 days ago | The World Trades Here: Bybit’s WSOT Is Coming Back With a Prize Pool of Up to 10 Million USDT event | 1 actors in common · 2 news media |
| 21 days ago | Bitget Wallet Brings USD and EUR Virtual Accounts to 100 Million Users event | 1 actors in common · 4 news media |