Perch AI Expands Free Model Access

// 14m agoPRODUCT UPDATE

Perch AI Expands Free Model Access

Perch AI’s free Starter plan includes up to $20 of monthly hosted-model usage with no credit card, including DeepSeek V4.1 Flash and GLM 5.3 Flash. Qwen 3.8 Max remains available through November 20, while weekend limits nearly double and DeepSeek reaches 456 tokens per second at p90.

// ANALYSIS

Perch is turning aggressive model pricing into a compelling developer on-ramp, but “free” still means metered access rather than unlimited inference.

–Starter includes 20,000 Perch Tokens, equivalent to up to $20 of hosted usage.
–DeepSeek’s 456-token/s figure is platform-reported p90 throughput across 16,400 promo replies, not an independent benchmark.
–The unlimited DeepSeek weekend offer applies to Pro and Ultra; free users receive expanded limits instead.
–Qwen 3.8 Max is a time-limited promotion ending November 20.
–Starter users should review Perch’s optional data program before sending sensitive code or documents.

// TAGS

perch-aillminferencepricingdevtoolhosted-serviceagent

DISCOVERED

14m ago

2026-11-07

PUBLISHED

39m ago

2026-11-07

RELEVANCE

8/ 10

AUTHOR

0x0SojalSec

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

LAUNCH47m ago

Result launches Mac beta for open models

Result’s Mac beta unifies local model downloads, cloud-hosted models, and private inference endpoints such as Ollama or DGX systems in one interface. Its broader vision is personal, sovereign AI with local execution and zero data retention.

RESEARCH56m ago

PSI Claims Lean-Checked Quantum Conjecture Proof

PSI says an internal research system, powered by a non-public Anthropic model, produced a proof of the Aaronson–Ambainis conjecture, an open problem since 2008. The formalization was accepted by Lean, while independent experts are still reviewing the paper and its connection to the original conjecture.

INFRA1h ago

DeepSeek V4.1 Flash Hits AI Gateway

Vercel AI Gateway now offers DeepSeek V4.1 Flash 1017 Ultrafast through Fireworks, with a 1M-token context window, multimodal input, reasoning, tool calling, and structured outputs. The endpoint targets over 1,000 tokens per second, though live throughput varies.