Search “Reasoning”

Meta Releases 816B Open-Source Model Muse Spark Open, Partnering with Apple for Local Execution

Meta has released the open-weight model Muse Spark Open, which uses a MoE architecture with 256 experts, activates about 42B parameters per token, has a context window of 1 million tokens, and supports text, image, video, and audio input but outputs text only. Meta has partnered with Apple to run the approximately 530 GByte four-bit quantized version on two interconnected Mac Studio M5 Ultra machines, distributing computation via the MLX framework and Thunderbolt 5, achieving about 25 output tokens/second at an 8192-token context. Artificial Analysis tests show that it scores 39 on the Intelligence Index, making it the strongest American open-weight model currently available, but it trails MiMo-V2.6-Pro's 46 points and GLM-5.3's 45 points, and ranks last in four Agent tests. The model weights are available on Hugging Face in two versions: 16-bit at approximately 1.6 TByte and four-bit at approximately 530 GByte, with API pricing at $1.25 per million input tokens and $4.25 per million output tokens.

English translation is being prepared.

StepFun open-sources Step 5 Preview, a 600B-parameter MoE model that ranks second on the OpenRouter weekly chart

StepFun officially releases the Step 5 Preview model weights, with BF16 and FP8 versions launched simultaneously on Hugging Face and ModelScope. The model uses a sparse MoE architecture, with 600B total parameters and 27B activated parameters, supports a 1 million Token context window, and scores 44 on Artificial Analysis's Intelligence Index, placing it in the global top three open-source models. As of October 15, 2026, its Token usage ranks second on both the OpenRouter weekly and daily charts, with about 18.8 trillion Tokens used in the past seven days; according to evaluation data at release, its average cost per task is $0.71, about 1/8 that of Claude Opus 5.

English translation is being prepared.

Swiss AI releases the Apertus 1.5 multimodal model, expanding context to 262k tokens

Swiss AI has released Apertus 1.5, which builds on 1.0 by adding native image and audio understanding, a thinking mode, and a 262,144-token context window—4 times larger than 1.0. The model uses early fusion to convert images and audio into discrete tokens and feed them into the same sequence as text; the architecture is unchanged, and it only understands multimodal input while outputting text only. The team says the 8B version outperforms open-source models of the same scale on instruction following and multilingual tasks, its multimodal performance is close to Gemma 4 12B, and it has also released the training data mixture, preference data, and intermediate checkpoints from each stage.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

Meta releases Muse Wonder and Muse Wonder Full, surpassing Claude on long-horizon coding tests…

Meta Superintelligence Labs launched its new-generation flagship models Muse Wonder and Muse Wonder Full on October 13. The Full version scored 56 on the Artificial Analysis Composite Intelligence Index, 2 points behind Claude Opus 5.5's 58. On the DeepSWE 1.1 long-horizon software engineering test, Wonder Full scored 79.0, surpassing Opus 5.5's 74.2 and GPT-6 Astra's 74.1, but it still trails Opus on Terminal-Bench 4.0 and ProgramBench. Artificial Analysis estimates its average task cost at $4.52, up 183% from the previous generation Muse Spark 1.3; a developer comparison on the same prompt shows its 3D game generation is noticeably worse than Opus, and Meta's model card has acknowledged 3D tasks as a known limitation.

English translation is being prepared.

Google Releases Gemini 4 Argon, Topping Multiple Leaderboards and Focusing on Cybersecurity Defense

Google has released its new-generation flagship model Gemini 4 Argon, which scores 77.9% on DeepSWE v1.1, surpassing Claude Opus 5.5's 74.2% and GPT-6 Astra's 74.1%, and tops the Vals Index with 68.90%. The model supports output of up to one million tokens, and during the promotional period costs $2 per million input tokens and $10 per million output tokens, with a per-question cost of about 40% of Astra's. It is currently available only to vetted cybersecurity defense teams, with the first batch entering Google's Fairwind program.

English translation is being prepared.

Anthropic releases Claude Sonnet 5.5, 30% faster than Sonnet 5 with unchanged per-token pricing

Anthropic has released Claude Sonnet 5.5, the second model in the Claude 5.5 family. It is smarter and more efficient than Sonnet 5 and 30% faster, with unchanged per-token pricing; because it needs fewer tokens to complete the same work, costs for most tasks drop by up to 30%. The official recommendation is to use Sonnet 5.5 for well-scoped everyday tasks such as bug fixing and rapid feature iteration, while complex long-running work should still use Opus 5.5; Haiku 5.5 will join the family in the coming weeks. The model ID is claude-sonnet-5-5, and it is already available on platforms including the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry. Starting with v2.1.284, the sonnet alias in Claude Code points to this model, which runs at medium effort by default and natively supports a 1M context window.

GPT-6 Family's First Model Astra Released, Author's Hands-On Test Consumed 4 Billion Tokens

OpenAI released Astra, which the author calls the beginning of the GPT-6 family; it performs outstandingly on ARC-AGI-3 and scored highest on Zapier's AutomationBench, with pricing the same as Fable 5.1. During weekend trials, the author consumed 4 billion tokens and found its capability fluctuates noticeably—it is a showcase model rather than one that reliably does work. Astra in Codex can now continue advancing tasks without waiting for an answer, and OpenAI also claims it has achieved its automated research intern goal, with the next goal being an automated AI researcher by March 2028.

English translation is being prepared.

Ant Ling releases finance-enhanced model Ling-3.0-flash-Fin

Ant Ling releases its first finance-enhanced model, Ling-3.0-flash-Fin, which continues the Ling-3.0-flash architecture, maintaining 124B total parameters and 5.1B activated parameters. Through continued pre-training on financial corpora, domain post-training, and tool-use optimization, it strengthens the handling of complex financial content such as annual reports and financial workbooks. The model was tested on benchmarks including FinFIRST, FinSearchComp Verified, Finance Agent, APEX-Agents, SpreadsheetBench, and τ³-Banking, with its AA Intelligence Index v4.1.1 score rising from 38 to 41. The model will offer a one-month limited-time free API on OpenRouter, and its weights will be open-sourced next week.

Ant Ling Open-Sources Ling-3.0-tiny-base and Ling-3.0-flash-base

Ant Ling has officially open-sourced two Base Models, Ling-3.0-tiny-base and Ling-3.0-flash-base, and simultaneously released the weights from three stages—pre-training, mid-training, and WSM merging—for a total of 6 checkpoints. Ling-3.0-tiny-base has 7.9B total parameters and 1.3B activated parameters, while Ling-3.0-flash-base has 124B total parameters and 5.1B activated parameters; both can be used for continued pre-training, supervised fine-tuning, preference optimization, reinforcement learning post-training, and distillation. The official notes state that the Base Models have not completed instruction alignment and are not recommended for direct deployment as end-user-facing chat services without post-training.