Sakana Namazu

Sakana AI's official blog index explicitly documents a Namazu productization milestone, stating that the company released Sakana Chat, a chat service powered by the Namazu α-series. This is a first-party confirmation that Sakana AI is the creator of the Namazu model line and that an α version is already being deployed Because the blog index is the supplied surface, the excerpt does not expose full release dates, parameter counts, or detailed API specifications for the α variant. The page nonetheless serves as the strongest primary-source anchor in the candidate set for Sakana Namazu's authorship and product status, and points to oth

Sakana Chat, launched in March 2026 by Japan-based Sakana AI, runs on "Namazu," a large language model tuned for the Japanese language and Japanese business practices, and pairs the model with web search so responses can draw on up-to-date information rather than only the model's internal knowledge. Namazu is not train On August 3, 2026, Sakana AI released an API version of the model branded "Sakana Namazu," using an OpenAI-compatible request format and shipping built-in web search and code execution tools, with pricing listed on Sakana AI's official pricing page rather than bundled into a flat subscription. The web service remains f

Sakana AI released "Sakana Namazu" as a developer API on August 3, 2026, making the Japan-focused model previously available only through the free Sakana Chat web service accessible from external applications, internal systems, and scheduled jobs. The v1.0 API release is distinct from the March 2026 alpha prototype tha The Namazu API is positioned for enterprise developer use cases such as AI agents, business automation, customer service platforms, and industry-specific applications that require natural Japanese phrasing and understanding of domestic systems and conventions rather than translated-from-English output. Differentiation

OminiGate's benchmark page for Sakana Namazu indexes Sakana's self-published evaluation results attributed to console.sakana.ai. It lists AIME26 at 96.67, MMLU-Pro at 90.33, LiveCodeBench v6 at 90.33, JFBench at 37.4, Translation at 52.2, and FairPoliticsQA at 56.3 as the single recorded result per metric. The page provides a clean, structured summary of Sakana's benchmark claims across both general reasoning and Japanese-specific evaluations. The page does not introduce independent measurement or methodology details, serving solely as an aggregator of Sakana's console-published numbers. It corroborates that Sakana Namazu's headline claim is the FairPoliticsQA improvement alongside maintenance of the Kimi K2.6 base's reasoning benchmarks. Limited additional analytical value beyond confirming the consistency of Sakana's published figures across third-party hosts.

Digital Applied's August 11, 2026 analysis frames Sakana Namazu as a fine-tune of Moonshot AI's open-weight Kimi K2.6 — explicitly by Sakana's own repeated statements — rather than a model Sakana pretrained from scratch. The piece reports Sakana's own FairPoliticsQA result of 56.30% versus 34.10% on the Kimi K2.6 base Reading the Namazu launch alongside the August 10 announcement that Sakana retrained the Fugu orchestration conductor on a Gemma 4 base, the article characterizes Sakana's broader strategy as "sovereignty by tuning and orchestration, not pretraining." This interpretation is analytical and secondary, not first-party, bu

AI Weekly reported on Sakana AI's release of the Sakana Namazu API, describing it as an OpenAI-compatible endpoint for the LLM previously running inside Sakana Chat, built on Moonshot AI's open Kimi K2.6 and tuned in-house for Japanese language and business norms. The article highlights integration ergonomics — teams on OpenAI SDKs can swap the base URL and API key — and notes built-in web search and code execution as the model's distinguishing tools. The piece relays Sakana's self-reported benchmark numbers, including the FairPoliticsQA jump from 34.10% to 56.30% versus the base model, alongside gains on JFBench, AIME26, and coding tasks. It flags that these are company-published figures without disclosed methodology and notes Sakana's post omits pricing, context window, and latency details. The article frames the release as part of a broader pattern of regional teams fine-tuning strong open-weight bases for local markets.

SakuttoAI's explainer confirms that Sakana AI opened the Sakana Namazu API on August 3, 2026, accessible the same day from the API Console, with the model identifier sakana-namazu-v1.0 served via https://api.sakana.ai/v1. The article details the Kimi K2.6 open-model foundation and Sakana's additional in-house training for Japanese language, Japanese business conventions, reduction of unnecessary refusals, and output bias mitigation. Built-in web search and code execution tools are confirmed, as is the EU/EEA/UK/Switzerland availability exclusion pending GDPR and regional regulatory compliance. The explainer frames Sakana Namazu as the model powering Sakana Chat, now productized for direct API integration into enterprise applications and internal systems. It emphasizes the OpenAI-compatible drop-in design, requiring only a base URL swap for existing OpenAI SDK code, and highlights target use cases such as internal documents and email processing in Japanese where correct honorific register and business conventions matter. Pricing is not addressed in the excerpt.

Sakana AI announced Sakana Namazu, a Japanese-specific large language model API designed to give enterprises generative AI capabilities tailored to Japan's language, business environment, and business practices. The API leverages the open-source Kimi K2.6 foundation model created by Moonshot AI and is additionally tune Namazu addresses gaps left by LLMs trained primarily on English-language data, where Japanese companies need models that understand honorific language usage, business etiquette, legal jargon, and industrial documentation. Sakana's approach merges the capabilities of Kimi K2.6 with added optimization aimed at Japanese-l

According to the LLM Stats leaderboard, Sakana Namazu ranks 49th overall on the composite LLM Stats Score and sits in the top 10% for Legal (5 of 211), Finance (6 of 228), and Healthcare (7 of 244). Its blended price is reported at $1.10 per million tokens on a log-scale cost-efficiency chart, positioned between GPT OS The LLM Stats benchmark scorecard for Sakana Namazu lists evaluated results across multiple datasets sourced from official Sakana performance charts versus the Kimi K2.6 base. Notable reported scores include AIME 2026 at 0.97/1 (rank 3) and LiveCodeBench v6 at 0.90/1 (rank 4), alongside MMLU-Pro coverage reflecting the

ModelBench's listing for Sakana Namazu records the model as a Japanese-specialized reasoning model built on Kimi K2.6, closed-weights, released August 3, 2026. It documents a 262.144K token context window, 65.536K max output, input price of $0.95/M tokens, output price of $4/M tokens, and cache read at $0.15/M tokens. The page lists reasoning, tool calling, structured output, attachments, and vision input as supported capabilities, with no open weights. The page identifies four inference providers — OpenRouter, Kilo Gateway, Merge Gateway, and Sakana AI itself — all listing identical $0.95 input rates, making price a non-differentiator among them. The model ID is confirmed as sakana-namazu with the family sakana-namazu and creator Sakana, and the API record links back to Sakana's own documentation. License and knowledge cutoff are noted as undocumented.

LLMBase's model listing for Sakana Namazu confirms it as a Japanese-specialized reasoning model from Sakana AI, based on Moonshot AI's Kimi K2.6 with additional training for Japanese language and business contexts. The listing records a 262,144-token context window, text-only modality, input price of $0.95 per 1M tokens, and output price of $4.00 per 1M tokens. The release date is listed as August 11, 2026, conflicting with Sakana's own August 3 announcement. The page notes that Sakana Namazu is not yet available in LLMBase's Chat or EU-hosted Inference API, and lists similar Sakana models in the Fugu family for context. It describes the model's suitability for Japanese instruction-following and everyday tasks including writing, summarizing, and brainstorming. As an aggregator, it provides no independent benchmark reproduction, relying on Sakana's published figures.

Benchgen's model card describes Sakana Namazu as Sakana AI's Japanese-specialized LLM API, released August 3, 2026, built by fine-tuning Moonshot AI's open-weight Kimi K2.6 on proprietary in-house Japanese business-context data. It documents a 262,144-token context window and pricing of $0.95/$4.00 per 1M input/output tokens with $0.15 per 1M cached input tokens, served through a drop-in OpenAI-compatible API with built-in web search and code execution tools. The card notes the model is not currently available in the EU, EEA, UK, or Switzerland pending regulatory compliance. The card lists Sakana's published benchmark results — 96.67% AIME26, 90.33% MMLU-Pro, 90.33% LiveCodeBench v6, 37.40% JFBench, 52.20% Japanese-English translation, and 56.30% FairPoliticsQA (up from the base model's 34.10%) — all attributed to console.sakana.ai without independent reproduction. It highlights Sakana Namazu's strengths in Japanese instruction-following and politically-neutral, non-US-centric answering while flagging the closed-weights nature of the API as a limitation versus the Apache 2.0-licensed Kimi K2.6 base.