←Search results

Search “Agents”

By relevance · Total 561 results Clear search

Wednesday

Monday

Tuesday

  5. 36Kr · AI

    Memory Tensor Open-Sources MemT 1.0 Series of Memory-Native Models

    Memory Tensor has open-sourced the MemT 1.0 series of memory-native models, comprising two variants: Nano with 36B parameters and 3B activated, and Flash with 288B parameters and 17B activated, built on the Qwen3.6-35B-A3B and DeepSeek-V4-Flash bases, respectively. The models add a memory branch parallel to attention within selected Transformer modules, encoding experience into an updatable matrix state that participates in subsequent generation. Across 500 SWE-bench Verified tasks, Nano achieved an average success rate of 69.73% at key checkpoints, versus a no-experience baseline of 66.07%; under a 128K long history, end-to-end inference speed is approximately 1.86 times that of the full-context approach, and the persistent state footprint per trajectory is one sixty-sixth that of the full-context KV Cache.

    Memory Tensor has open-sourced the MemT 1.0 series of memory-native models, comprising two variants: Nano with 36B parameters and 3B activated, and Flash with 288B parameters and 17B activated, built on the Qwen3.6-35B-A3B and DeepSeek-V4-Flash bases, respectively. The models add a memory branch parallel to attention within selected Transformer modules, encoding experience into an updatable matrix state that participates in subsequent generation. Across 500 SWE-bench Verified tasks, Nano achieve

    Matched fieldsBody

    #hot-signal#Model releases#Open-source ecosystem#English text is being prepared.
  6. Unite.AISelected

    Inception releases flagship diffusion large model Mercury 3

    Inception released the diffusion-architecture large model Mercury 3 on October 19, 2026, claiming it surpasses GPT-6 Luna and Claude Haiku 5.5 on multiple benchmarks, reaches a speed of 1068 tokens/s on NVIDIA GPUs, and increases the context window from 260,000 in Mercury 2.5 to 1 million tokens. The launch price is $0.10 per million input tokens and $0.35 for output, with regular pricing at $0.20 and $0.70, and cached input at $0.02. Mercury 3 can be accessed via the Inception API, OpenRouter, Baseten, and Vercel, and is compatible with the OpenAI API; Mercury 2 and 2.5 are no longer available to new users, and existing customers can use them until November 15, 2026.

    Inception released the diffusion-architecture large model Mercury 3 on October 19, 2026, claiming it surpasses GPT-6 Luna and Claude Haiku 5.5 on multiple benchmarks, reaches a speed of 1068 tokens/s on NVIDIA GPUs, and increases the context window from 260,000 in Mercury 2.5 to 1 million tokens. The launch price is $0.10 per million input tokens and $0.35 for output, with regular pricing at $0.20 and $0.70, and cached input at $0.02. Mercury 3 can be accessed via the Inception API, OpenRouter,

    Matched fieldsBody

    #hot-signal#Model releases#Reasoning#Agents
  10. Fireworks AI

    Meta's open-weight model Muse Spark Open launches on Fireworks: 816B parameters, 1M context

    Meta's open-weight model Muse Spark Open is available on Fireworks from day one of launch. It is an 816B-parameter MoE model that activates 42B parameters per token, with a 1M-token context window, targeting long-horizon coding and agentic tasks. Fireworks says it ranks first among 75 models on Vals AI's Harvey legal agent benchmark with 25.4%, nearly 4 times that of Claude Opus 5 (6.7%); on the Rogo Big Finance Bench it surpasses Claude Opus 5.5 at every difficulty level. Fireworks validated the weights with its own production stack, scoring 91.1% on AIME26, 89.7% on GPQA Diamond, and 84.3% on SWE-bench Verified, and reproduced consistent results with SGLang.

    e on Fireworks from day one of launch. It is an 816B-parameter MoE model that activates 42B parameters per token, with a 1M-token context window, targeting long-horizon coding and agentic tasks. Fireworks says it ranks first among 75 models on Vals AI's Harvey legal agent benchmark with 25.4%, nearly 4 times that of Claude Opus 5 (6.7%); on the Rogo Big Finance Bench it surpasses Claude Opus 5.5 at every difficulty level. Fireworks validated the weights with its own production stack, scoring 91.

    Matched fieldsBody

    #editorial#Product updates#Model releases#English text is being prepared.

Sunday

Saturday

  1. ZhidongxiSelected

    DeepSeek V4.1-Flash updated to V4.1-Flash-1017 and opens search and fetch APIs

    DeepSeek V4.1-Flash has been updated to version V4.1-Flash-1017; the official web, App and API have been updated in sync, and the model weights are available for download on HuggingFace and ModelScope. In the DeepSWE v1.1 test, the new model's max tier achieved a single-task pass rate of 77.0%, 2.8 percentage points higher than the previous version, and the average cost per task dropped from about 3.3 yuan to about 1.3 yuan. DeepSeek also opened to the public the Search and Fetch APIs of its self-developed search engine; both interfaces are priced at 5 yuan per thousand calls, and this part of access via server-side tool calls will be charged per call starting from 00:00 Beijing time on November 1.

    DeepSeek V4.1-Flash has been updated to version V4.1-Flash-1017; the official web, App and API have been updated in sync, and the model weights are available for download on HuggingFace and ModelScope. In the DeepSWE v1.1 test, the new model's max tier achieved a single-task pass rate of 77.0%, 2.8 percentage points higher than the previous version, and the average cost per task dropped from about 3.3 yuan to about 1.3 yuan. DeepSeek also opened to the public the Search and Fetch APIs of its sel

    Matched fieldsBody

    #hot-signal#Model releases#Open-source ecosystem#English text is being prepared.