XMH.NET — AI industry & image API insights https://xmh.net/blog/ AI industry milestones, image model updates, and practical analysis for business and development teams. en Sat, 10 Oct 2026 12:50:20 +0800 SynthID Detector expands the conversation about AI content provenance https://xmh.net/blog/synthid-detector-2026/ Google's October 2026 access expansion made verification tools more visible to publishers and creative teams. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/synthid-detector-2026/ Claude Haiku 5.5 adds adjustable effort to high-volume work https://xmh.net/blog/claude-haiku-55-2026/ Anthropic's October 2026 release brought more configuration choices to its small-model tier. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/claude-haiku-55-2026/ The Decisions API gives classification a more explicit interface https://xmh.net/blog/openai-decisions-api-2026/ OpenAI's October 2026 beta separates typed decisions from open-ended generated responses. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/openai-decisions-api-2026/ Gemini 4 Argon's staged release separates capability from availability https://xmh.net/blog/gemini-4-argon-2026/ Google's September 2026 announcement emphasized long-running work while beginning with a restricted rollout. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/gemini-4-argon-2026/ GPT-6.1 Sol makes endpoint compatibility part of the cost comparison https://xmh.net/blog/openai-gpt-61-sol-2026/ OpenAI's September 2026 release offered another option for complex work, with an important distinction between API routes. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/openai-gpt-61-sol-2026/ Claude Sonnet 5.5 puts token efficiency beside response speed https://xmh.net/blog/claude-sonnet-55-2026/ Anthropic's September 2026 release emphasized everyday professional tasks and lower work-level costs in its model comparisons. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/claude-sonnet-55-2026/ Anthropic's Life Sciences Verification Program formalizes specialized access https://xmh.net/blog/claude-life-sciences-2026/ The September 2026 announcement showed how frontier-model access can depend on the organization and intended use, not only an API key. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/claude-life-sciences-2026/ AlphaGenome Atlas turns model predictions into a searchable research resource https://xmh.net/blog/alphagenome-atlas-2026/ Google's September 2026 announcement highlighted the value of precomputed outputs and accessible interfaces for scientific AI. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/alphagenome-atlas-2026/ WeatherNext 3 shows how AI models can become a data service https://xmh.net/blog/weathernext-3-2026/ Google's September 2026 weather release emphasized frequent updates, live observations, and usable forecast outputs. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/weathernext-3-2026/ GPT-6 Astra raises the bar for evaluating demanding AI work https://xmh.net/blog/openai-gpt-6-astra-2026/ OpenAI's September 2026 release emphasized complex reasoning and professional tasks, with specific integration requirements. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/openai-gpt-6-astra-2026/ GPT-5.6 introduces Sol, Terra, and Luna as explicit model tiers https://xmh.net/blog/openai-gpt-56-2026/ OpenAI's July 2026 family release made tier naming and routing more visible parts of application architecture. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/openai-gpt-56-2026/ Claude Sonnet 5 emphasizes sustained execution in an everyday model tier https://xmh.net/blog/claude-sonnet-5-2026/ Anthropic's June 2026 release brought more attention to agents that plan, use tools, and complete multi-step work. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/claude-sonnet-5-2026/ Claude Science puts reproducible artifacts at the center of an AI workbench https://xmh.net/blog/claude-science-2026/ Anthropic's June 2026 beta connected research tools with an environment designed to preserve how outputs were produced. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/claude-science-2026/ Gemma 4 QAT puts model compression under a quality microscope https://xmh.net/blog/gemma-4-qat-2026/ Google's June 2026 checkpoints illustrated how training-aware compression can change local deployment options. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/gemma-4-qat-2026/ Gemma 4 12B explores a simpler multimodal architecture for laptops https://xmh.net/blog/gemma-4-12b-2026/ Google's June 2026 release emphasized local operation and a unified approach to visual and audio inputs. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/gemma-4-12b-2026/ GPT-5.5 shows why model upgrades need configuration review https://xmh.net/blog/openai-gpt-55-2026/ OpenAI's April 2026 API release brought another professional-work model with behavior that could differ from an existing integration's assumptions. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/openai-gpt-55-2026/ GPT-5.4 links professional work with a larger tool surface https://xmh.net/blog/openai-gpt-54-2026/ OpenAI's March 2026 release made long context and tool-enabled workflows part of the model migration discussion. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/openai-gpt-54-2026/ Gemini 3.1 Flash-Lite targets the economics of frequent requests https://xmh.net/blog/gemini-31-flash-lite-2026/ Google's March 2026 preview put high-volume, latency-sensitive workloads at the center of a smaller model tier. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/gemini-31-flash-lite-2026/ Gemini 3.1 Pro focuses the comparison on complex problem solving https://xmh.net/blog/gemini-31-pro-2026/ Google's February 2026 release strengthened its reasoning-oriented model lineup across developer and consumer products. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/gemini-31-pro-2026/ Claude Sonnet 4.6 makes model-tier boundaries worth revisiting https://xmh.net/blog/claude-sonnet-46-2026/ The February 2026 release emphasized coding, computer use, and long-context work in Anthropic's middle tier. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/claude-sonnet-46-2026/ Claude Opus 4.6 brings larger project context into demanding work https://xmh.net/blog/claude-opus-46-2026/ Anthropic's February 2026 release emphasized sustained coding and professional tasks, with a larger context option in beta. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/claude-opus-46-2026/ Xcode's Claude Agent SDK integration closes a visual feedback loop https://xmh.net/blog/claude-xcode-2026/ The February 2026 integration brought more autonomous coding work into Apple's native development environment. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/claude-xcode-2026/ NVIDIA Rubin makes AI infrastructure a system-level discussion https://xmh.net/blog/nvidia-rubin-2026/ The January 2026 platform announcement combined compute, networking, and memory-related components in a coordinated design. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/nvidia-rubin-2026/ GPT-5.2 makes context management part of professional AI work https://xmh.net/blog/openai-gpt-52-2025/ OpenAI's December 2025 release put more attention on sustained tasks, reasoning configuration, and complex deliverables. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/openai-gpt-52-2025/ Mistral 3 gives open-model teams a wider size range https://xmh.net/blog/mistral-3-2025/ The December 2025 release paired a large mixture-of-experts model with smaller models under a permissive license. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/mistral-3-2025/ DeepSeek-V3.2 combines reasoning with tool use https://xmh.net/blog/deepseek-v32-2025/ The December 2025 release made mode support and endpoint lifetime important details for developers evaluating agent workflows. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/deepseek-v32-2025/ FLUX.2 brings reference consistency into the image-production conversation https://xmh.net/blog/flux-2-2025/ Black Forest Labs' November 2025 release emphasized editing and controlled creative workflows rather than isolated visual samples. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/flux-2-2025/ Claude Opus 4.5 puts cost per completed task ahead of token price https://xmh.net/blog/claude-opus-45-2025/ Anthropic's November 2025 release emphasized demanding professional work, making evaluation scope especially important. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/claude-opus-45-2025/ Gemini 3 broadens the expectations placed on multimodal models https://xmh.net/blog/gemini-3-pro-2025/ Google's November 2025 launch linked stronger reasoning with developer tools and consumer experiences. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/gemini-3-pro-2025/ Claude Haiku 4.5 makes fast specialist work more attractive https://xmh.net/blog/claude-haiku-45-2025/ Anthropic's October 2025 release strengthened the case for assigning routine subtasks to a smaller model. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/claude-haiku-45-2025/ Gemini Computer Use turns visual understanding into interface actions https://xmh.net/blog/gemini-computer-use-2025/ Google's October 2025 preview extended Gemini 2.5 capabilities to agents operating graphical interfaces. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/gemini-computer-use-2025/ DeepSeek-V3.2-Exp brings sparse attention to the efficiency debate https://xmh.net/blog/deepseek-v32-exp-2025/ The September 2025 experimental release focused on reducing the computation involved in long-context processing. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/deepseek-v32-exp-2025/ Claude Sonnet 4.5 makes the agent runtime part of the release story https://xmh.net/blog/claude-sonnet-45-2025/ Anthropic's September 2025 launch combined model improvements with tools for longer and more recoverable workflows. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/claude-sonnet-45-2025/ GPT-5 makes reasoning effort a configurable product tradeoff https://xmh.net/blog/openai-gpt-5-2025/ OpenAI's August 2025 API family gave developers several model sizes and an explicit way to control deliberation. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/openai-gpt-5-2025/ Qwen3-Coder connects open coding models with terminal workflows https://xmh.net/blog/qwen-3-coder-2025/ The July 2025 release paired a specialized model with Qwen Code, emphasizing repository-scale work and tool use. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/qwen-3-coder-2025/ Magistral brings multilingual reasoning into enterprise evaluations https://xmh.net/blog/mistral-magistral-2025/ Mistral's June 2025 reasoning-model announcement highlighted a requirement that English-only benchmarks can miss. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/mistral-magistral-2025/ Apple opens its on-device foundation model to app developers https://xmh.net/blog/apple-foundation-models-2025/ The June 2025 framework announcement offered a new path for integrating AI into native applications with offline operation in mind. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/apple-foundation-models-2025/ Claude 4 puts long-running agent work in the spotlight https://xmh.net/blog/claude-4-2025/ Anthropic's May 2025 Opus and Sonnet release emphasized sustained execution, coding, and reasoning with tools. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/claude-4-2025/ Imagen 4 and Veo 3 raise expectations for creative production tools https://xmh.net/blog/veo-3-imagen-4-2025/ Google's May 2025 media announcements highlighted the difference between a model release and a complete production workflow. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/veo-3-imagen-4-2025/ Qwen3 makes thinking mode and model size explicit deployment choices https://xmh.net/blog/qwen-3-2025/ The April 2025 release combined dense and mixture-of-experts models with different approaches to inference-time reasoning. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/qwen-3-2025/ o3 and o4-mini bring visual reasoning into a broader model comparison https://xmh.net/blog/openai-o3-o4-mini-2025/ OpenAI's April 2025 reasoning releases encouraged teams to evaluate complex tasks that combine text, code, and images. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/openai-o3-o4-mini-2025/ GPT-4.1 shows why instruction following deserves its own test suite https://xmh.net/blog/openai-gpt-41-2025/ OpenAI's April 2025 family release emphasized coding, long context, and more reliable adherence to developer instructions. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/openai-gpt-41-2025/ Agent2Agent introduces a protocol for cooperation across AI systems https://xmh.net/blog/a2a-2025/ Google's April 2025 A2A announcement addressed how agents discover capabilities and coordinate work across application boundaries. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/a2a-2025/ Llama 4 brings native multimodality to a new open-weight generation https://xmh.net/blog/llama-4-2025/ Meta's April 2025 Scout and Maverick release made architecture, context, and deployment configuration inseparable evaluation questions. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/llama-4-2025/ Gemini 2.5 Pro makes thinking behavior part of integration planning https://xmh.net/blog/gemini-25-pro-2025/ Google's March 2025 experimental release put reasoning at the center of a new Gemini generation. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/gemini-25-pro-2025/ Blackwell Ultra links AI infrastructure to longer reasoning workloads https://xmh.net/blog/nvidia-blackwell-ultra-2025/ NVIDIA's March 2025 announcement emphasized inference-time computation and the systems needed to support it. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/nvidia-blackwell-ultra-2025/ Gemma 3 brings multimodal options to smaller deployment footprints https://xmh.net/blog/gemma-3-2025/ Google's March 2025 family release widened the choices for teams evaluating local and single-accelerator AI. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/gemma-3-2025/ The Responses API makes tools part of the application interface https://xmh.net/blog/openai-responses-api-2025/ OpenAI's March 2025 release gave developers a new foundation for model interactions that involve search, files, and actions. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/openai-responses-api-2025/ Claude 3.7 Sonnet and Claude Code connect reasoning with execution https://xmh.net/blog/claude-37-2025/ Anthropic's February 2025 launch paired a hybrid reasoning model with a terminal-based coding research preview. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/claude-37-2025/ o3-mini brings reasoning into more cost-sensitive workflows https://xmh.net/blog/openai-o3-mini-2025/ OpenAI's January 2025 release gave developers a smaller reasoning model with familiar application-building features. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/openai-o3-mini-2025/ Qwen2.5-VL puts document layout and visual grounding in focus https://xmh.net/blog/qwen-25-vl-2025/ The January 2025 announcement expanded the role of vision-language models beyond recognizing objects in photographs. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/qwen-25-vl-2025/ DeepSeek-R1 makes open reasoning models a serious evaluation topic https://xmh.net/blog/deepseek-r1-2025/ The January 2025 release combined reasoning-focused training, published model materials, and smaller distilled variants. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/deepseek-r1-2025/ DeepSeek-V3 adds momentum to efficient open-model development https://xmh.net/blog/deepseek-v3-2024/ The December 2024 release put mixture-of-experts design and deployment economics back into enterprise comparisons. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/deepseek-v3-2024/ Gemini 2.0 signals a shift toward models that use tools https://xmh.net/blog/gemini-2-2024/ Google's December 2024 announcement connected multimodal capabilities with a broader vision for agentic applications. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/gemini-2-2024/ MCP standardizes a connection point for AI tools and data https://xmh.net/blog/mcp-2024/ Anthropic's November 2024 protocol announcement addressed the integration work around models rather than the models themselves. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/mcp-2024/ Claude's computer-use beta changes the boundary of automation https://xmh.net/blog/claude-computer-use-2024/ Anthropic's October 2024 release let developers experiment with agents that observe screens and interact with software interfaces. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/claude-computer-use-2024/ The Realtime API makes conversation a state-management problem https://xmh.net/blog/openai-realtime-api-2024/ OpenAI's October 2024 preview showed why voice interfaces need more than a fast text response. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/openai-realtime-api-2024/ Llama 3.2 separates vision workloads from edge deployment https://xmh.net/blog/llama-32-2024/ Meta's September 2024 release combined image-understanding models with smaller text models intended for local devices. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/llama-32-2024/ Qwen2.5 widens the choice of general and specialized open models https://xmh.net/blog/qwen-25-2024/ The September 2024 release highlighted the importance of matching a checkpoint, its license, and its specialization to the application. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/qwen-25-2024/ o1-preview puts reasoning effort on the engineering roadmap https://xmh.net/blog/openai-o1-preview-2024/ OpenAI's September 2024 preview brought a different performance tradeoff to complex problem-solving tasks. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/openai-o1-preview-2024/ Structured Outputs turns formatting into an API contract https://xmh.net/blog/openai-structured-outputs-2024/ OpenAI's August 2024 feature helped developers request schema-constrained responses while leaving factual validation to the application. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/openai-structured-outputs-2024/ FLUX.1 gives image teams another model family to evaluate https://xmh.net/blog/flux-1-2024/ Black Forest Labs' August 2024 launch expanded the image-generation landscape with distinct hosted and downloadable variants. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/flux-1-2024/ The EU AI Act enters into force: why an AI inventory became useful https://xmh.net/blog/eu-ai-act-2024/ The August 2024 milestone gave organizations a reason to document where and how AI was being used. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/eu-ai-act-2024/ Llama 3.1 expands the scope of open-weight enterprise experiments https://xmh.net/blog/llama-31-2024/ Meta's July 2024 release combined a 405B model with longer context and a broader deployment ecosystem. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/llama-31-2024/ GPT-4o mini makes small-model routing worth testing https://xmh.net/blog/openai-gpt-4o-mini-2024/ The July 2024 release added a lower-cost option for routine text and image-understanding tasks. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/openai-gpt-4o-mini-2024/ Gemma 2 makes hardware fit a central part of open-model selection https://xmh.net/blog/gemma-2-2024/ Google's June 2024 release offered 9B and 27B models designed with inference efficiency in mind. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/gemma-2-2024/ Claude 3.5 Sonnet raises the value of a middle-tier model https://xmh.net/blog/claude-35-sonnet-2024/ The June 2024 release strengthened the case for measuring useful work rather than assuming that model size determines quality. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/claude-35-sonnet-2024/ Apple Intelligence brings AI into the operating-system conversation https://xmh.net/blog/apple-intelligence-2024/ Apple's June 2024 announcement emphasized personal context, device integration, and a mix of local and cloud processing. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/apple-intelligence-2024/ Gemini 1.5 Flash puts response speed into the model-selection debate https://xmh.net/blog/gemini-15-flash-2024/ Google's May 2024 launch emphasized a lighter option for developers handling frequent, well-defined requests. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/gemini-15-flash-2024/ GPT-4o makes multimodal input a mainstream API consideration https://xmh.net/blog/openai-gpt-4o-2024/ The May 2024 release encouraged developers to build workflows that understand images as well as written instructions. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/openai-gpt-4o-2024/ AlphaFold 3 shows why specialized AI deserves separate evaluation https://xmh.net/blog/alphafold-3-2024/ The May 2024 molecular modeling announcement broadened the AI conversation beyond general-purpose assistants. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/alphafold-3-2024/ Llama 3 brings open-weight models further into everyday development https://xmh.net/blog/meta-llama-3-2024/ Meta's April 2024 launch widened the practical options for teams building and operating their own AI applications. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/meta-llama-3-2024/ DBRX puts mixture-of-experts deployment on enterprise roadmaps https://xmh.net/blog/databricks-dbrx-2024/ Databricks' March 2024 release showed why model architecture and operating requirements should be evaluated together. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/databricks-dbrx-2024/ NVIDIA Blackwell shifts attention to the whole AI serving system https://xmh.net/blog/nvidia-blackwell-2024/ The March 2024 platform announcement made infrastructure efficiency central to the economics of AI deployment. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/nvidia-blackwell-2024/ Claude 3 makes model tiers a practical architecture choice https://xmh.net/blog/anthropic-claude-3-2024/ Haiku, Sonnet, and Opus presented developers with explicit tradeoffs between speed, capability, and cost in March 2024. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/anthropic-claude-3-2024/ Mistral Large expands the enterprise model shortlist https://xmh.net/blog/mistral-large-2024/ The February 2024 release added another major supplier to enterprise AI evaluations and highlighted distribution as a buying consideration. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/mistral-large-2024/ Gemma's first release gives developers another deployment choice https://xmh.net/blog/google-gemma-2024/ Google's 2B and 7B open models made local experimentation easier, while leaving operational responsibility with the deploying team. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/google-gemma-2024/ Gemini 1.5 Pro and the practical limits of long context https://xmh.net/blog/google-gemini-15-2024/ Google's February 2024 announcement made million-token context a major industry topic. Useful retrieval still requires careful evaluation. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/google-gemini-15-2024/ Bard becomes Gemini: understanding Google's product and model naming https://xmh.net/blog/google-gemini-advanced-2024/ The February 2024 rebrand brought Google's consumer assistant and model family under one name, with implications for software buyers. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/google-gemini-advanced-2024/ OpenAI embedding v3: why retrieval quality became a product decision https://xmh.net/blog/openai-embeddings-v3-2024/ OpenAI's January 2024 embedding update put multilingual search and vector design on the same roadmap as model selection. Sat, 10 Oct 2026 12:50:20 +0800 AI Industry https://xmh.net/blog/openai-embeddings-v3-2024/ Nano Banana 2.1: evaluate the next step in Gemini image workflows https://xmh.net/blog/nano-banana-2-1-release/ The latest Nano Banana release and a practical review of text, repeated edits, and wide compositions. Sat, 10 Oct 2026 12:00:00 +0800 Gemini / Nano Banana https://xmh.net/blog/nano-banana-2-1-release/ GPT Image 2.5 Sunburst and Flare: choose around the image job https://xmh.net/blog/gpt-image-2-5-sunburst-flare/ A practical evaluation framework for the two GPT Image 2.5 model options. Sat, 10 Oct 2026 12:00:00 +0800 GPT Image https://xmh.net/blog/gpt-image-2-5-sunburst-flare/ Transparent backgrounds in GPT Image 2: review the edges https://xmh.net/blog/gpt-image-2-transparent-backgrounds/ The August 2026 transparency preview and a checklist for assets that need an alpha channel. Sat, 10 Oct 2026 12:00:00 +0800 GPT Image https://xmh.net/blog/gpt-image-2-transparent-backgrounds/ Nano Banana 2 Lite: evaluate a leaner image production path https://xmh.net/blog/nano-banana-2-lite-release/ How to test Gemini 3.1 Flash Lite Image against the requirements of high-volume creative work. Sat, 10 Oct 2026 12:00:00 +0800 Gemini / Nano Banana https://xmh.net/blog/nano-banana-2-lite-release/ Older GPT Image models: plan ahead for December 1, 2026 https://xmh.net/blog/legacy-gpt-image-model-retirements/ A coordinated migration for GPT-image-1-mini, GPT Image 1.5, and chatgpt-image-latest. Sat, 10 Oct 2026 12:00:00 +0800 GPT Image https://xmh.net/blog/legacy-gpt-image-model-retirements/ Nano Banana 2 and Pro reach GA: move from preview to stable IDs https://xmh.net/blog/nano-banana-2-pro-general-availability/ What to verify when adopting gemini-3.1-flash-image and gemini-3-pro-image. Sat, 10 Oct 2026 12:00:00 +0800 Gemini / Nano Banana https://xmh.net/blog/nano-banana-2-pro-general-availability/ GPT-image-1 retirement: prepare for October 23, 2026 https://xmh.net/blog/gpt-image-1-retirement-plan/ A migration checklist for teams still using the original GPT-image-1 model. Sat, 10 Oct 2026 12:00:00 +0800 GPT Image https://xmh.net/blog/gpt-image-1-retirement-plan/ GPT Image 2: reassess output sizes before changing the default https://xmh.net/blog/gpt-image-2-release/ The April 2026 model release and a review of the image contract between your API and publishing tools. Sat, 10 Oct 2026 12:00:00 +0800 GPT Image https://xmh.net/blog/gpt-image-2-release/ Nano Banana 2: evaluating Gemini 3.1 Flash Image Preview https://xmh.net/blog/nano-banana-2-preview/ A practical pilot plan for the second-generation Nano Banana release. Sat, 10 Oct 2026 12:00:00 +0800 Gemini / Nano Banana https://xmh.net/blog/nano-banana-2-preview/ GPT Image in the Batch API: plan for completion, not instant previews https://xmh.net/blog/gpt-image-batch-api/ How to evaluate asynchronous image work for catalogs and other scheduled production tasks. Sat, 10 Oct 2026 12:00:00 +0800 GPT Image https://xmh.net/blog/gpt-image-batch-api/ JSON image edits: choosing URLs, file IDs, or multipart uploads https://xmh.net/blog/openai-image-edits-json-inputs/ The February 2026 Images API update and what it changes in an image asset pipeline. Sat, 10 Oct 2026 12:00:00 +0800 GPT Image https://xmh.net/blog/openai-image-edits-json-inputs/ Gemini 2.5 Flash Image Preview has shut down: clean up old references https://xmh.net/blog/gemini-flash-image-preview-shutdown/ A post-retirement check for applications, examples, and saved image jobs. Sat, 10 Oct 2026 12:00:00 +0800 Gemini / Nano Banana https://xmh.net/blog/gemini-flash-image-preview-shutdown/ A GPT Image edit fix: why explicit settings need regression checks https://xmh.net/blog/gpt-image-low-fidelity-edit-fix/ The January 2026 fidelity correction and a small test strategy for image-editing applications. Sat, 10 Oct 2026 12:00:00 +0800 GPT Image https://xmh.net/blog/gpt-image-low-fidelity-edit-fix/ GPT Image 1.5 in the Responses API: choosing an integration surface https://xmh.net/blog/gpt-image-1-5-responses-tool/ Why an orchestration feature and a direct image endpoint serve different application needs. Sat, 10 Oct 2026 12:00:00 +0800 GPT Image https://xmh.net/blog/gpt-image-1-5-responses-tool/ GPT Image 1.5: review an upgrade without losing your baseline https://xmh.net/blog/gpt-image-1-5-release/ A controlled evaluation for the December 2025 GPT Image model release. Sat, 10 Oct 2026 12:00:00 +0800 GPT Image https://xmh.net/blog/gpt-image-1-5-release/ Gemini 2.5 Flash Image Preview: prepare for the retirement date https://xmh.net/blog/gemini-flash-image-preview-retirement-notice/ What to inventory when a preview identifier receives a shutdown notice. Sat, 10 Oct 2026 12:00:00 +0800 Gemini / Nano Banana https://xmh.net/blog/gemini-flash-image-preview-retirement-notice/ Nano Banana Pro: a review plan for more demanding image briefs https://xmh.net/blog/nano-banana-pro-preview/ The Gemini 3 Pro Image preview and a structured way to evaluate complex marketing and design requests. Sat, 10 Oct 2026 12:00:00 +0800 Gemini / Nano Banana https://xmh.net/blog/nano-banana-pro-preview/ Nano Banana input tokens fall: what changes in an editing budget? https://xmh.net/blog/nano-banana-image-input-token-update/ The 1,290-to-258 image-input token change and how to separate input savings from the rest of an image request. Sat, 10 Oct 2026 12:00:00 +0800 Gemini / Nano Banana https://xmh.net/blog/nano-banana-image-input-token-update/ GPT-image-1-mini: evaluating a lower-cost image workflow https://xmh.net/blog/gpt-image-1-mini-cost-evaluation/ Where a smaller image model can help, and why cost per approved asset matters more than request price alone. Sat, 10 Oct 2026 12:00:00 +0800 GPT Image https://xmh.net/blog/gpt-image-1-mini-cost-evaluation/ Gemini 2.5 Flash Image moves to general availability https://xmh.net/blog/gemini-2-5-flash-image-ga/ How to turn a successful image preview experiment into a versioned application integration. Sat, 10 Oct 2026 12:00:00 +0800 Gemini / Nano Banana https://xmh.net/blog/gemini-2-5-flash-image-ga/ The original Nano Banana: Gemini 2.5 Flash Image enters preview https://xmh.net/blog/nano-banana-first-generation/ The first Nano Banana release, viewed through a practical reference-image editing workflow. Sat, 10 Oct 2026 12:00:00 +0800 Gemini / Nano Banana https://xmh.net/blog/nano-banana-first-generation/ Preserving product details with GPT-image-1 input fidelity https://xmh.net/blog/gpt-image-1-high-input-fidelity/ A practical reading of the July 2025 high-fidelity guide for edits where logos, packaging, and identity matter. Sat, 10 Oct 2026 12:00:00 +0800 GPT Image https://xmh.net/blog/gpt-image-1-high-input-fidelity/ GPT-image-1: bringing image generation into a production workflow https://xmh.net/blog/gpt-image-1-api-launch/ What the April 2025 API release changed for product imagery, and how to evaluate generation and editing before rollout. Sat, 10 Oct 2026 12:00:00 +0800 GPT Image https://xmh.net/blog/gpt-image-1-api-launch/