daily notes

DeepSeek moved from API update to open weights

DeepSeek turned the morning's API-only update into deployable infrastructure by publishing V4 Flash 0731 weights under MIT. Its model card reports Terminal-Bench 2.1 at 82.7 versus 61.8 for April's preview, while vLLM says the million-token DSpark serving configuration carries over.

@ArtificialAnlysoriginal post ↗

DeepSeek V4 Flash 0731 scores 50 on the Artificial Analysis Intelligence Index, a 10-point jump over DeepSeek V4 Flash (released April 2026) that puts it 6 points ahead of DeepSeek V4 Pro. It shares identical architecture and pricing with the earlier DeepSeek V4 Flash, and lands on our Pareto frontier for Intelligence vs Cost per Task @deepseek_ai's DeepSeek V4 Flash 0731 is one Intelligence Index point behind GPT-5.6 Luna (max, 51). Even after OpenAI's 80% price cut on GPT-5.6 Luna today, DeepSeek V4 Flash 0731's Cost per Task on DeepSeek's first-party API comes in at ~60% lower than GPT-5.6 Luna (max), a model with comparable intelligence. A key driver of this is DeepSeek's ~98% cache hit discount on its first-party API, a significantly more aggressive discount than the 90% cache hit discount offered by most of the industry The new model is a significant step up from the previous generation, DeepSeek V4 Flash (40), and places the model within 1 point of GLM-5.2 (max, 51). It remains 7 points behind the open weights frontier set by Kimi K3 (max, 57). For additional context, this places the model in line with recently released Gemini 3.6 Flash (50) and 1 point behind Muse Spark 1.1 (xhigh, 51). DeepSeek is expected to release the model's full weights in the coming weeks DeepSeek V4 Flash 0731 retains a 1M token context window, and its size remains unchanged from DeepSeek V4 Flash at 284B total parameters and 13B active at inference time Key results: - Improvements in agentic performance: DeepSeek V4 Flash 0731 achieves an Elo rating of 1559 on GDPval-AA v2, up from 1189 for the previous DeepSeek V4 Flash. Terminal-Bench 2.1 rises 17 points to 79% and tau3-Bench Banking 8 points to 31%. - Token usage falls 12% against the predecessor: about 206M output tokens versus about 234M for the previous model across the Intelligence Index. - AA-Omniscience improvements are driven by fewer hallucinations rather than higher accuracy. Its hallucination rate is 84%, a 12-point decrease, while accuracy is unchanged. Pricing remains $0.14/$0.28 per 1M input/output tokens, with a $0.0028 cache-hit price.

@ArtificialAnlysoriginal post ↗

DeepSeek V4 Flash 0731 is now open weights! @deepseek_ai has just released the weights for its new flash tier model, DeepSeek V4 Flash 0731. With a score of 50 on the Artificial Analysis Intelligence Index, it lands among the top 3 open weights models on the leaderboard. The weights are released under the MIT license, allowing unrestricted commercial use and modification. DeepSeek V4 Flash 0731 shares identical architecture and pricing with the earlier DeepSeek V4 Flash. At a size of 284B total parameters (13B active), released in mixed FP4/FP8 precision at ~167GB total file size, it lands on our Pareto frontier for Intelligence Index vs. Total Parameters. Among open weights models, DeepSeek V4 Flash 0731 delivers a significant leap in intelligence for its size class. DeepSeek V4 Flash 0731 is also available now through DeepSeek's first-party API. Check out Artificial Analysis to compare DeepSeek V4 Flash 0731 with other leading open weights and proprietary models: https://t.co/zeUIrzHIOC

@vllm_projectoriginal post ↗

🎉 Congrats to @deepseek_ai on open-sourcing the official DeepSeek-V4-Flash-0731 release! A sparse MoE with 256 routed experts and six active per token, a 1M-token context window, and three reasoning-effort levels. Built for code agents and tool use. 🤖 Same architecture as the DSpark preview checkpoint, so your serving config carries over. And the DSpark draft module ships inside these weights, so speculative decoding is one flag instead of a second model to load: --speculative-config '{"method":"dspark","num_speculative_tokens":7,"draft_sample_method":"greedy"}' Model: https://t.co/nH83pfMdfw

Evaluation security became a cross-lab control problem

A second lab turned evaluation escape from a one-off model incident into an infrastructure-control problem. Anthropic found three real compromises across 141,006 runs after a third-party range exposed internet access; the continuing review leaves details open, but sealed connectivity and least privilege are now cross-lab requirements.

@AnthropicAIoriginal post ↗

In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different organizations. Our post describes what happened, how it happened, and what we're changing. We encourage other AI developers to perform similar reviews. We conducted this review together with @Irregular, one of our evaluation partners, and thank them for the joint investigation and their collaboration on this post. This type of collaboration is increasingly critical to safe, rigorous evaluation of models, and we look forward to continuing to work together on security.

Video editing became a first-class model task

MiniMax H3 made instruction-based video editing a first-class generation task. MiniMax's API accepts text, image, video, and audio references for 2K clips lasting four to fifteen seconds, while Artificial Analysis ranked it first for video editing and Krea framed it as a lower-cost Seedance alternative.

@ArtificialAnlysoriginal post ↗

MiniMax H3 is #1 in Video Editing and ranks top 3 in both Text to Video and Image to Video on Artificial Analysis. MiniMax plans to release its weights under the MiniMax Community License, which would make it the leading open weights model by far. The model accepts text, images, videos, and audio clips as input to generate 5-15 second clips at 24fps with native audio generation. Video input support also enables instruction-based video editing, where it takes the #1 spot in our Video Editing Leaderboard. The model is also #2 in Text to Video and #3 in Image to Video. MiniMax H3 is available now in the Hailuo AI app and via the MiniMax API. MiniMax lists the 2K tier at $0.13 per second, or $7.80 per minute, with a 768p tier marked as coming soon.

@krea_aioriginal post ↗

introducing MiniMax H3. this new model matches seedance's features at a fraction of the cost, and it will soon be open-weights!

discussion3 selected replies

@krea_aireply ↗

1/ product editorials turn product shots into editorials. H3 handles typography, brand assets, and motion. up to 2k, with native stereo audio on every generation. https://t.co/AI73wQsVyw

@krea_aireply ↗

2/ audio sync feed H3 music or a dialogue clip and the video syncs to it. https://t.co/l9il4qlCii

@krea_aireply ↗

3/ first frame / last frame upload a first frame, a last frame, or both, and h3 generates everything in between. perfect for seamless transitions. https://t.co/ztC6yiqxhK

Labor costs accelerated while layoffs stayed low

Layoffs stayed low while labor costs accelerated. June civilian compensation rose 0.9% quarter over quarter, with benefits up 1.0%, while initial claims remained 197,000 and their four-week average fell to 202,750; private-industry real wages still declined 0.4% over the year.