CMOtech US - Technology news for CMOs & marketing decision-makers
United States
Featherless adds MiniMax M3.1 Flash for coding tasks

Featherless adds MiniMax M3.1 Flash for coding tasks

Mon, 16th Nov 2026 (Today)
Mara Sugue
MARA SUGUE News Editor

Featherless has added MiniMax M3.1 Flash to its serverless inference platform. The model is aimed at coding and autonomous software tasks.

Developers can access MiniMax M3.1 Flash through Featherless via an API aligned with the OpenAI standard, with a 256K context window in FP4. The integration also includes vision support and a zero-logging policy for data privacy.

The model has 428 billion total parameters, with 23 billion active per token. MiniMax built it for software development, agentic workflows and computer use, focusing on resolving software engineering tasks in repositories, handling business processes across tools and applications, operating desktop software from on-screen information, and identifying and reproducing vulnerabilities in codebases.

Model changes

MiniMax M3.1 Flash retains the mixture-of-experts design used in M3, but changes the serving approach in three areas. Sparse attention now runs across the full model rather than only after the first three full-attention layers, as in M3.

MiniMax also quantised experts and attention paths during training, with experts at 4-bit precision and the rest at 8-bit. It added a DSpark draft head to increase generation speed through speculative decoding.

Before the formal release, an earlier version of the model operated anonymously under the name Space Bunny. It topped public usage leaderboards and served more than 200 trillion tokens before its identity became known, according to Featherless.

Cost and ranking

On Artificial Analysis' Intelligence Index, MiniMax M3.1 Flash scored 29, compared with 19 for MiniMax M3. Featherless said the cost of running the full evaluation fell to USD $0.49 per task from USD $1.29 for M3.

That puts the model one point above GLM-5.3-Flash on the index at a lower cost per task, while trailing DeepSeek V4.1 Flash 1017, which scored 32. In hard reasoning, the model reached 48% on Humanity's Last Exam, up from 39% for M3, according to Featherless.

The launch follows Featherless' introduction of Simple Jev, an open-source library designed to turn open-source AI models into zero-shot classification engines.

Commenting on the release, Eugene Cheah, Chief Executive Officer and Co-Founder of Featherless, said: "Engineering teams require powerful models that are economical enough to perform autonomous operations 24/7. By scoring 29 on the Intelligence Index while costing only $0.49 per task, M3.1 Flash offers exactly that. These workloads are very data-heavy, and organisations normally face rapidly escalating per-token bills. That's why we're offering M3.1 Flash with a fully tuned inference stack on dedicated hardware with fixed monthly pricing. This way, businesses and engineering teams can scale their workflows without unpredictable expenses."