Kimi K3.1 Pro

MarkTechPost·Oct 23, 2026

Moonshot AI Releases Kimi K3.1 Pro: A 2.8T Total/104B Active Params (MoE) Model with 1M Context

Moonshot AI has released Kimi K3.1 Pro, the first model in its K3.1 series. This open-weight model features 2.8 trillion total parameters and 104 billion active parameters using a Mixture of Experts (MoE) architecture, with a 1-million-token context window. The Kimi K3.1 Pro is positioned to compete with strong proprietary models on coding, agentic, and knowledge benchmarks. While open weights are anticipated for release in November 2026, API access is available now. The model is a post-trained version of Kimi K3, enhanced with significantly larger post-training compute and joint reinforcement learning across numerous domains. The Moonshot team optimized its RL framework, long-context inference, sandbox infrastructure, and rollout scheduling. The base K3 architecture incorporates Kimi Delta Attention (KDA) and Attention Residuals (AttnRes), activating 16 out of 896 experts via a Stable LatentMoE framework. The API supports text, image, and video inputs. Kimi K3.1 Pro Highspeed, a faster version, runs 2 to 3 times quicker. On benchmarks, K3.1 Pro scores 65.7% on Terminal-Bench 4.0, closely trailing Claude Opus 5.5 at 66.4%. It leads on SWE-Marathon with 51.7%, surpassing Opus 5.5's 46.3%.

The Decoder·Oct 23, 2026

Chinese AI model Kimi K3.1 Pro is closing in on Anthropic's Claude Opus 5.5

The Hacker News·Oct 1, 2026

OpenAI Disrupts Reasoning Extraction Campaign Linked to Moonshot AI Associates

OpenAI announced on Wednesday that it has successfully identified and disrupted a coordinated campaign aimed at illicitly extracting protected reasoning from its artificial intelligence (AI) models. The company stated that a "core cluster of the activity," which began in the first week of July, has been attributed to individuals associated with Moonshot AI, a Chinese AI company headquartered in Beijing. This campaign represents a significant effort to bypass security measures and gain unauthorized access to the proprietary reasoning processes embedded within OpenAI's advanced AI systems. The disruption of this campaign highlights the ongoing challenges in safeguarding AI model integrity and intellectual property. AI models, particularly large language models (LLMs) like those developed by OpenAI, are trained on vast datasets and involve complex architectures and reasoning capabilities that represent substantial research and development investments. The extraction of this "protected reasoning" could allow competitors or malicious actors to replicate or reverse-engineer these capabilities without undertaking the extensive resources required for independent development. This practice, often referred to as "model distillation" or "model extraction," can lead to the creation of unauthorized or counterfeit AI models that may lack the safety and ethical guardrails implemented by the original developers. OpenAI has not provided specific details regarding the technical methods used in the extraction campaign or the exact nature of the "protected reasoning" being targeted. However, the attribution to Moonshot AI suggests a potential international dimension to these efforts. Moonshot AI, as a Beijing-based entity, operates within a rapidly evolving and competitive global AI landscape. The company's involvement, as alleged by OpenAI, points to the strategic importance of AI model capabilities and the lengths to which some entities might go to acquire them. OpenAI's proactive disruption of this campaign underscores its commitment to protecting its intellectual property and maintaining the security and integrity of its AI technologies. The company's statement implies that further details may be forthcoming as its investigation progresses, but the immediate focus has been on neutralizing the threat and informing the public about the nature of the activity. This incident also raises broader questions about AI security and the governance of AI development globally. As AI capabilities become increasingly sophisticated and integrated into various sectors, the protection of the underlying models and their reasoning processes becomes paramount. The ability to extract and replicate these capabilities could have significant implications for market competition, national security, and the responsible deployment of AI. OpenAI's action serves as a warning to other AI developers and a signal to the broader AI community about the persistent threats to AI model security. The company's efforts to combat such campaigns are crucial for fostering trust and ensuring the continued innovation and ethical development of artificial intelligence technologies.

MarkTechPost·Sep 28, 2026

Fireworks AI Releases Ember-1: A Post-Trained Kimi K3 That Uses About 40% Fewer Tokens

Fireworks AI has released Ember-1, a specialized model developed by Fireworks Research through post-training Moonshot AI's open-weight Kimi K3. This new model is designed to produce shorter reasoning traces while preserving task accuracy, a feat achieved by fundamentally altering the model's reasoning process rather than simply adjusting inference settings. According to a release post from Fireworks, Ember-1 delivers the quality of Kimi K3 with approximately 40% fewer tokens. Currently, Ember-1 is available as a Research Preview exclusively through the Fireworks serverless API. Fireworks has not released the model's weights, training code, or specific training algorithms, making self-hosting impossible at this time. The development of Ember-1 addresses a significant challenge with current reasoning models, where they can expend over 90% of generated tokens on internal reasoning processes. This inefficiency becomes particularly costly in multi-turn agentic workloads, where prior reasoning is replayed to the model in each subsequent turn. The context window grows roughly quadratically with the number of turns, leading to long reasoning traces from early turns being re-read and re-billed on every later call. Fireworks stated that customers desired Kimi K3's coding capabilities at a reduced cost, but lowering K3's reasoning effort setting resulted in an unacceptable loss of quality. Consequently, the Fireworks team focused on training the model to reason more efficiently. Fireworks Research constructed Ember-1 by retaining useful self-reflection behaviors, such as revisiting assumptions or responding to feedback, while eliminating redundant reasoning and unproductive loops. The training dataset encompasses a broad range of tasks, including mathematics, coding, instruction following, conversation, search, tool use, and software engineering. This dataset includes both standalone problems and extended multi-step interactions. The training process utilized task and environment feedback to guide on-policy planning and learning. The Fireworks team conducted over 50 training experiments and more than 200 evaluations. They also developed new, unpublished training algorithms. All training operations were performed on Fireworks' own infrastructure. This innovation by Fireworks AI aims to make advanced reasoning models more cost-effective for applications that require extensive sequential processing or agentic behavior. By reducing the token overhead associated with internal thought processes, Ember-1 could enable more complex and longer-running AI agent tasks without a proportional increase in computational cost. The focus on maintaining accuracy while reducing token count is a critical step towards more efficient and scalable AI deployments, particularly in competitive cloud API markets where cost per token is a key metric for adoption. The model's availability as a research preview suggests that Fireworks AI is seeking feedback and further refinement before a broader commercial release.

MarkTechPost·Sep 12, 2026

Cognition Releases SWE-2: A Kimi K3 Post-Trained Coding Model That Matches Fable 5.1 on FrontierCode at 64% Lower Cost

Cognition, the developer of the Devin AI coding agent, has launched SWE-2, its latest and most advanced coding model. This new model is post-trained using reinforcement learning techniques applied to Kimi K3, an open-source large language model developed by Moonshot AI, which boasts 2.8 trillion parameters. Cognition reports that SWE-2 achieved a score of 50.0% on the FrontierCode 1.1 Main benchmark. This performance is notably close to Fable 5.1, a leading coding model, with only a 1-point difference, while operating at a 64% lower cost. SWE-2 also introduces selectable reasoning-effort levels, a feature developed within a single reinforcement learning run, allowing users to adjust the model's computational intensity and performance trade-offs. SWE-2 is built upon the foundational infrastructure and methodology established by its predecessor, SWE-1.7, which was similarly post-trained but utilized Kimi K2.7 as its base model. For SWE-2, Cognition scaled its reinforcement learning process to accommodate models with nearly three times the parameters of the previous iteration. The company states that its reinforcement learning approach continues to find significant performance improvements, even on top of the Kimi K3 model, adding an estimated 5 to 6 percentage points to scores across various benchmarks. A key innovation in SWE-2 is its reinforcement learning algorithm, which enables the training of multiple effort levels simultaneously within a single training session. Each effort level is associated with a specific cost penalty, allowing the model to optimize its cost-performance frontier holistically. In terms of benchmark performance, SWE-2 demonstrates strong results across several evaluations. On FrontierCode 1.1 Main, it scored 50.0%, surpassing Kimi K3's 44.2% and Grok's 48.0%. It also achieved 73.0% on DeepSWE 1.1, outperforming Kimi K3 (68.5%) and Fable 5.1 (67.4%). On Terminal-Bench 2.1, SWE-2 reached an impressive 92.8%, exceeding Kimi K3 (88.3%) and Fable 5.1 (91.4%). Cognition claims that SWE-2 comes within a few percentage points of GPT-6 Astra's performance but at approximately one-quarter of the cost. However, the model shows a comparative weakness on Terminal-Bench 4, where it scored 27.3%, lagging behind Fable 5.1 (55.8%) and GPT-6 Astra (57.9%) by a significant margin. It is important to note that FrontierCode is a proprietary benchmark developed by Cognition, and the reported scores for rival models were also generated by Cognition's internal evaluations. SWE-2 is currently not available for independent deployment on users' own infrastructure, as it does not offer open weights or a standalone API. Access to SWE-2 is exclusively through Cognition's Devin platform, including Devin: Desktop and CLI versions, with Devin Web and Fusion interfaces slated for future rollout. This integrated approach ensures that the model's capabilities are utilized within Cognition's ecosystem.