Starlink V5의 Cybercab 통합·motif 13B/314B MoE 공개·Gemini Batch p95 -80% 등 최신 X 트렌드

Starlink V5 directly integrated in Cybercab

elie

@eliebakouch

huge open weight release, motif (korean company) just released a 13B active 314B total MoE performing on par with bigger models like minimax M3 and deepseek v4 Pro they incorporate their own research bets with per expert activation function (polynorm) and a variant of differential attention (they say GDLA, my guess is gated differential latent attention?) as well as modified mHC

Logan Kilpatrick

@OfficialLoganK

We just landed some big infra upgrades for the Gemini Batch API: - p95 latency decreased by 80% -p99 latency decreased by 68% - batch success rate is now >99.998% - 98% reduction in batch expirations - added support for partial batches great work by the team to land this!!

elvis

@omarsar0

Huge launch from @tryramp . Different steps in an agent workflow can use different models. This can help reduce costs significantly without sacrificing performance. Model routing will become a core part of the stack. Now available in one OpenAI-compatible endpoint.

Claude Code 2.1.216 is now available. 40 CLI changes, 1 system prompt change Highlights: • Added sandbox.filesystem.disabled setting to skip filesystem isolation while keeping network egress control • Fixed quadratic message normalization in long sessions, preventing multi-second stalls on resume • Resumed background agent sessions restore agent prompt and tool restrictions to preserve configured behavior Complete details in thread ↓

Claude Code 2.1.216 additional updates CLI surface: Added: • models: claude-code-guide Removed: • models: claude-code-guideG File: https:// github.com/marckrenn/clau de-code-changelog/commit/2d38a31d86c9b3e01d19d2c4b1fd404ffff04ec8/meta/cli-surface.md#diff-662031a066e433468319e799350331e143e4635468b9c2924019d16654027e31L938-R938 … Metadata: • Time since 2.1.215 release: 1d 18h 31m 5s • Bundle file size: +332.7 kB (+0.9%) • Prompt files: +2 (+13.3%) • Prompt tokens: +3,099 (+14.0%) Prompt token mix by kind: • system: 38.4%→52.3% • tools: 61.6%→47.7% Files: • metadata: https:// github.com/marckrenn/clau de-code-changelog/commit/2d38a31d86c9b3e01d19d2c4b1fd404ffff04ec8/meta/metadata.md#diff-a08dfa163baef97ea4f56d29009ac65f5f7dbc46e3a58e80d7aa06a2a2520899L10-R10 … • prompt stats: https:// github.com/marckrenn/clau de-code-changelog/commit/2d38a31d86c9b3e01d19d2c4b1fd404ffff04ec8/meta/prompt-stats.md#diff-fa8e31336ece71d159f37b6fd4ed0c61939849140bd5c08c7ca793b5b2ae7f58L1-R1 … Other prompt/string highlights: 1) Documents that the write tool writes files to the local filesystem. Links: 1) https:// github.com/marckrenn/clau de-code-changelog/commit/51f2448f3c9df75aadcb209ea4d6a2f159d91134/system-prompts/tool-description-write.md#diff-9c34cc0e7d9c17b756de147507de09c530bbab89236f20357009c4ab54a14f9aR18 … More details: https:// github.com/marckrenn/clau de-code-changelog/releases/tag/v2.1.216 …

OpenAI Developers

@OpenAIDevs

Sites is now available in the UK, EEA, and Switzerland for Plus and Pro accounts

Paige Bailey

@DynamicWebPaige

You can now fine-tune models on your personal laptops, and run the latest @GoogleGemma 4 models with as little as 3GB VRAM (). Stellar work, as always, from the @UnslothAI team!