---
title: 'In the News: October 19, 2026 (Extra 2)'
description: "MiniMax's M3.1 Flash lists $0.15 input and $0.60 output per million tokens and claims 73.7% on DeepSWE v1.1; all figures are the vendor's own."
canonical_url: 'https://darkfactory.dev/news/2026-10-19-extra-2'
markdown_url: 'https://darkfactory.dev/news/2026-10-19-extra-2.md'
collection: news
date_published: '2026-10-19T16:05:00-04:00'
date_modified: '2026-10-19T16:05:00-04:00'
---

# In the News: October 19, 2026 (Extra 2)


MiniMax released M3.1 Flash today at $0.15 per million input tokens and $0.60 per million output tokens. Every performance figure comes from MiniMax, and the first people to comment who had used the stealth version were not impressed.

## 1. MiniMax releases M3.1 Flash at $0.15 and $0.60 per million tokens, with vendor-reported coding scores

**[MiniMax M3.1 Flash: Redefining Agent Performance and Cost](https://www.minimax.io/blog/minimax-m31f)** · MiniMax Research · minimax.io, October 19, 2026

MiniMax says M3.1 Flash costs 70% less to use on agentic tasks than its predecessor, which it does not name in the post. Listed prices are $0.15 input, $0.60 output and $0.015 cached input per million tokens, so cached input is one tenth of the input rate. For coding, MiniMax reports 73.7% on DeepSWE v1.1, and 61.7 on OSWorld V2 for computer use. It also claims output speeds above 200 tokens per second, with a median above 170. The model is available through the MiniMax API, its M Plan subscription and MiniMax Code. Weights are promised "later". We read the post's text, not its charts.

The post says the model was tested anonymously beforehand as "Space Bunny", which it reports handled more than 200 trillion tokens. On the [Hacker News thread](https://news.ycombinator.com/item?id=50154540) (231 points and 85 comments about three hours after posting), a commenter wrote that the stealth version "failed to impress anyone as far as I can tell" and called the launch benchmarks "pretty benchmaxed". Another said that when they used Space Bunny through free providers, it was a "you get what you pay for" experience. Both are single-user reports, not measurements. A third commenter pointed out that the post compares against Opus 5 and GPT-5.6 Sol rather than the labs' newest models; we did not check the comparison charts.

**Why it matters:** Cached input at a tenth of the input rate is the line item that matters for agent loops, which resend a long stable prefix every turn. It makes M3.1 Flash worth pricing against your current cheap backend, but the scores are MiniMax's own and the only hands-on reports we found were negative. Run it on your own tasks, with your own cache-hit pattern, before moving any traffic.