model releaseinclusionAI

inclusionAI releases Ling 3.1 Flash: 560B MoE, 25B active, 262K context, free on OpenRouter

TL;DR

inclusionAI has released Ling 3.1 Flash, a hybrid reasoning mixture-of-experts model with 560B total and 25B active parameters and a 262K-token context window. It is listed as free on OpenRouter through NovitaAI. No benchmark scores have been published on the listing.

3 min read
0

Ling 3.1 Flash at a glance

Context window1000K tokens
Input$0.06/1M tokens
Output$0.18/1M tokens

How Ling 3.1 Flash ranks now

BenchmarkScoreRankSource
AA Intelligence Index29.6 index30 of 56Artificial Analysis, Nov 2026 ↗

Rank is against every model in our database with a sourced score on that benchmark, each counted once with its best score.

All Ling 3.1 Flash specs and scores →

inclusionAI has released Ling 3.1 Flash, a hybrid reasoning mixture-of-experts (MoE) model with 560B total parameters and 25B active per token. It has a 262K-token context window. According to its OpenRouter listing, the model was released on October 2, 2026 and is currently priced at $0 per 1M input tokens and $0 per 1M output tokens.

Key specifications

Spec Detail
Developer inclusionAI
Architecture Mixture-of-experts, hybrid reasoning
Total parameters 560B
Active parameters 25B
Context window 262K tokens
Release date Oct 2, 2026 (per OpenRouter listing)
Pricing Free (input and output)
Training cutoff Not disclosed
Benchmark scores Not published on the listing

Availability and performance

The model is available through OpenRouter, with NovitaAI as the sole listed provider. OpenRouter's listing shows these figures:

  • Latency: 1.12s P50
  • Throughput: 66 tokens per second P50
  • Provider uptime: 100.00% (NovitaAI)
  • OpenRouter availability: 98.93% over the last 24 hours

The listing does not say whether the free pricing is promotional or how long it will last. Paid rates have not been disclosed. The listing also does not itemize supported modalities. Its description refers to a language model, and inclusionAI sells vision capability separately in the Ling 3.0 Flash VL variant.

Position within the Ling family

Ling 3.1 Flash is much larger than earlier "Flash" models in the lineup, according to OpenRouter's catalog:

  • Ling 3.0 Flash: 124B total, about 5.1B active
  • Ling-2.6-flash: 104B total, 7.4B active
  • Ling 3.1 Flash: 560B total, 25B active

That is roughly 4.5x the total parameters and about 5x the active parameters of Ling 3.0 Flash. It sits below the trillion-parameter Ling-2.6-1T and Ring-2.6-1T (63B active) models. Unlike the Ling-2.6 instruct models, it combines instant and reasoning behavior in one hybrid model, as Ling 3.0 Tiny and Ling 3.0 Flash VL also do.

inclusionAI has not published benchmark results for 3.1 Flash on the listing, so its capability claims cannot yet be verified against peers.

What this means

The jump from 5B to 25B active parameters under the "Flash" name suggests inclusionAI is moving that tier toward higher capability, at the expense of the very low serving cost that defined Ling 3.0 Flash ($0.021 input / $0.063 output per 1M tokens). Active parameter count drives inference cost, so paid pricing for 3.1 Flash will probably be well above the 3.0 tier once it is set.

The free listing makes it easy to test now. A 66 tokens-per-second throughput on a single provider is moderate for a model with 25B active parameters. Builders should treat the free tier as an evaluation window rather than a production commitment until pricing and rate limits are published.

The main gap is evidence. Without benchmark scores, a technical report, or a stated training cutoff, the model's standing against other open MoE models at this scale is unknown. Independent evaluations on coding, tool use and long-context tasks will determine whether the larger active footprint is justified.

Related Articles

model release

Thinking Machines Lab releases Inkling 2 Mini: 26B-parameter looped MoE with 3B active and 1M-token context

Thinking Machines Lab has released Inkling 2 Mini, a multimodal Mixture-of-Experts model with about 26B physical total parameters and 3B active per token. It repeats its middle layers three times, supports a 1M-token context, and ships with a DSpark drafter. Hugging Face reports it leads 32 of 48 benchmarks against four comparable models.

model release

Tencent releases Hy4 flagship weights: 770B-parameter MoE, 1,048,576-token context, conditional license

Tencent released downloadable weights for its flagship Hy4 on November 20, with a System Card. The multimodal MoE has a 770-billion-parameter language backbone, 49 billion active parameters per text token and a 1,048,576-token context window, under a license with commercial conditions for model-as-a-service operators.

research

inclusionAI releases SingProbe, an 8.13M-parameter streaming guardrail for Ling-3.1-flash-1019

inclusionAI published Ling-3.1-flash-1019-singprobe, an 8.13M-parameter probe that reads three hidden-state layers of Ling-3.1-flash-1019 and scores query intent, response unsafety and hallucination risk at each generated token. The model card reports a response safety F1 of 0.8707 and a hallucination AUC of 0.8041.

model release

Tencent releases Hy4 flash: 220B MoE, 12B active parameters, 1M-token context

Tencent has released Hy4 flash, a multimodal Mixture-of-Experts model with 220B total parameters, 12B active per token, and a 1,048,576-token context window. Tencent calls it its most cost-effective Hy model to date and claims a per-task cost of 1/20 that of Hy4 on its internal WorkIndex benchmark.

Comments

Loading...