세일즈포스가 드림포스 2026에서 자사 첫 추론 모델인 Koa를 공개했습니다. 엔비디아의 오픈웨이트 모델인 Nemotron을 기반으로 만들어졌고, 영업과 마케팅, 고객 지원 업무에 특화된 것이 특징입니다. 지금까지 기업용 AI는 클로드나 챗지피티 같은 프론티어 모델에 의존하는 게

Mistral introduces Europe’s first AI reasoning model, Magistral, to challenge OpenAI and Google with open-source power.

YouTube video by Websplaining

French AI firm Mistral will release its first reasoning model Tuesday, taking on the likes of OpenAI and China’s DeepSeek.

Investing.com -- Alibaba (NYSE:BABA) has launched its most advanced reasoning model to date, the Qwen3-235B-A22B-Thinking-2507, following three months of significant enhancements to its Qwen3 technology. The new model features improved capabilities in logical reasoning, mathematics, science, and coding. It also demonstrates enhanced general skills including instruction following, tool use, and alignment. A key technical advancement in the new model is its 256K native context window, which allows for deeper, long-form understanding. The model is built specifically for thinking mode functionality and supports extended reasoning chains to maximize depth and accuracy, without requiring manual activation of thinking mode. Alibaba has made the new model available through both Qwen Chat and via API on Alibaba Cloud, allowing users to experience the technology directly. This article was generated with the support of AI and reviewed by an editor. For more information see our T&C.

It can be made even cheaper. PhD students from Berkeley have created an AI that supposedly is comparable to DeepSeek's R1-Zero model. Allegedly, the entire AI has been developed for a grand total of j...

Deep Think utilizes extended, parallel thinking and novel reinforcement learning techniques for significantly improved problem-solving.

Mistral AI unveils Magistral, a domain-specific reasoning model for enterprises. Enhanced logic, multilingual capabilities, and open-source options.

Chinese artificial intelligence startup DeepSeek released an update to its R1 reasoning model in the early hours of Thursday, stepping up competition with U.S. rivals such as OpenAI. Fiona Jones reports.

DeepSeek is looking to press home its advantage. The Hangzhou-based firm is accelerating the launch of the successor to January's R1 model.

This is for anyone who finds this and/or is still having this issue. Here is the mistake I made to cause this: export async function runResponseLoop({ client, model, tools, metadata, instructions, messages, availableTools, toolMetadata, reasoningModel, }: ResponseLoopParams): Promise<Response> { let response: Response; let previousResponseId: string | undefined; let loopMessages: ResponseInput = Array.isArray(messages) ? [...messages] : []; try { while (true) { response = await client.responses.create({ model, // reasoning model input: loopMessages, tools, ...(metadata && { metadata }), ...(instructions && { instructions }), ...(reasoningModel && previousResponseId ? { previous_response_id: previousResponseId } : {}), }); previousResponseId = response.id; const toolCalls = filterToolCalls(response); if (toolCalls.length === 0) { break; } const results = await executeToolCalls(toolCalls, availableTools, { metadata: toolMetadata, }); const nextMessages: ResponseInput = []; // This was causing the error - I was passing the output trigger for a tool call to the next loop. And because it was missing the reasoning item, it was failing with ‘function_call’ was provided without its required ‘reasoning’ item // toolCalls.forEach((tc) => nextMessages.push(tc)); // Only pass the tool call output to the next loop results.forEach((r) => nextMessages.push(r)); loopMessages = nextMessages; } return response; } catch (error) { throw error; } }

We introduce SELF-DISCOVER, a general framework for LLMs to self-discover the task-intrinsic reasoning structures to tackle complex reasoning problems that are challenging for typical prompting method...