Open Source Factory

MODELS & COSTS DeepSeek Flash-1017: Does the Update Reduce Rework? Open Source Factory · Oct 17, 2026 · 6 min read Watch the English motion graphic English editorial overview. Motion graphics and freshly composed original music: Open Source Factory. Sources: DeepSeek’s release notes and model card. Not a recording of a model test. DeepSeek changed Flash without changing its API rate. That gives existing users a straightforward reason to revisit a difficult job. The useful comparison is whether the October checkpoint needs fewer rescue prompts, failed calls and extra tokens to finish the same work. In this article What changed in the patch? Where do the published scores actually move? Will the same price produce a smaller bill? Why are there game demos and a search API in this release? What should existing users do with it? Retry a job that needed too much help The update is DeepSeek-V4.1-Flash-1017 , announced on October 17. DeepSeek says it performs better while finishing tasks...

MODELS & COSTS DeepSeek V4.1 Flash: Release Changes and the Pro Correction Open Source Factory · Sep 10, 2026 · 11 min read October 3, 2026 update: The V4-Pro retirement plan was reversed. Pro service and billing continue. The headline now distinguishes 8B active for input processing from 16B for output generation. The 748B total in the earlier title combined a 552B backbone with 196B of conditional memory; the body already explained that distinction. The official Flash release added architecture and pricing details to the temporary beta. It also left a migration trap in the early reporting: the proposed retirement of Pro later changed. This account keeps that service correction beside the release figures so a historical plan does not become current routing advice. In this article The September 10 release and later service correction Flash and Pro in the vendor tests The architectural changes The published rate changes Throughput and measurement conditions The reasoning control ...

MODELS & COSTS DeepSeek's September Beta: A Historical Test Window Open Source Factory · Sep 08, 2026 · 6 min read October 3, 2026 update: Historical beta coverage: the September 10 expiry has passed. Use the linked released model and current API history for availability; the quoted beta speeds are community reports, not this blog’s measurements. The endpoint name included its expiry date: deepseek-v4.1-flash-expires-on-0910. That short September window gave developers something to try before a full release, not a stable integration target. The reported 507 tok/s peak also belongs to uncontrolled beta observations rather than a matched speed test. In this article The September 8 temporary endpoint Developer-reported speed The architecture described for the beta Rates, tokens and the bill The Pro-replacement survey question The preceding release sequence Linked source documents At the time, the beta used V4 Flash rates and a 20-concurrent-request account cap. Developer repor...

MODELS & COSTS DeepSeek Flash Vision: API Access and Weights Are Separate Open Source Factory · Aug 31, 2026 · 5 min read An experimental API and a downloadable checkpoint create different ways to evaluate Flash Vision. The August 21 and August 31 dates refer to those separate releases. The model's text and multimodal scores remain DeepSeek's measurements until a matching independent run is available. In this article API and weight-release dates Text-agent results in the vendor table Multimodal-agent results Position within the V4 family API integration notes Architecture and deployment notes The gist: Same 284B / 13B-active Flash architecture with vision modules added and continued training. On text-only agents it stays on par with Flash 0731 (Terminal Bench 83.9 vs 82.7, DeepSWE 59.3 vs 54.4) — and on multimodal agents it jumps +10.3 on ApexBench to 36.5 , closing to within 2.9 of Opus 4.8. 1M context, MIT, 384 tokens per image at Flash pricing. The repository is d...

MODELS & COSTS DeepSeek Flash Vision: Read the Text Gains as a Comparison Open Source Factory · Aug 21, 2026 · 5 min read Several text-agent scores rose in DeepSeek's vision release even though those tasks used no images. That makes the vendor table interesting to inspect. It does not isolate vision training as the cause, because the released models and their training can differ in more than one way. In this article Vision and text-only Flash in the vendor table The Opus-4.8 comparison Experimental endpoint and inputs What the measurements establish The release is deepseek-v4-flash-vision-exp , an experimental multimodal variant of the Flash line. It accepts images alongside text for tasks such as screenshot reading and chart analysis. The experimental endpoint and the text model should be compared as distinct releases. on the text-only agent benchmarks — the ones with no images at all — the vision model beats the text-only Flash-0731 released three weeks ago. Not ties. Be...