SShortSingh.
TopTechnologyAI & MLStartupsProgrammingScienceBusinessFinanceCrypto & Web3World
Back to feed
Programming

DeepSeek Routes All V4-Pro API Traffic to Faster, Cheaper Flash Model

0
·1 views

DeepSeek began redirecting all requests sent to its flagship V4-Pro model to the newer V4.1-Flash starting September 14, 2026, while V4.1-Pro remains without a release date. The company justified the move by citing benchmark results showing V4.1-Flash outperforms V4-Pro on capability, speed, and cost across most agentic and coding tasks. V4.1-Flash also surpassed competing models from Anthropic and OpenAI on Terminal-Bench 2.1, despite being priced under one dollar per million tokens. However, the newer model showed measurable regressions in factual knowledge benchmarks, including a notable drop on SimpleQA from 55.2 to 42.3. The shift contrasts with rivals Z.ai, which maintains both flagship and Flash tiers, and Moonshot, which focuses solely on a single large-scale model.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

How Lastyard.space Built Zero-Loss Telemetry Infrastructure for Endurance Athletes

Lastyard.space has engineered a high-availability telemetry ingestion system that treats athlete workout data with the same rigor as production distributed infrastructure. The platform captures real-time metrics — including heart rate, cadence, elevation, and pace — from Apple Watch via a mobile edge collector, routing data through WireGuard tunnels and a Traefik ingress layer into a durable PostgreSQL event log. A core design principle is at-least-once delivery with idempotent writes, meaning duplicate or delayed samples from flaky mobile connections are merged rather than double-counted, using a composite primary key of athlete ID, workout ID, metric, and sample timestamp. The architecture separates the write ingestion path from live SSE read streams, ensuring dashboards and race control panels remain functional even under backend pressure. The system was validated against a real 21.1 km trail run logged on 12 September 2026, demonstrating stable data capture across 115 minutes of sustained metabolic load.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Apple folds Wi-Fi, cellular, and battery into one icon on iPhone Duo cover screen

Apple's upcoming iPhone Duo features a redesigned status bar that combines the three classic indicators — Wi-Fi, battery, and cellular — into a single small circle on its compact 5.4-inch cover screen. The design layers three visually distinct shapes: a continuous arc for battery, a fan symbol for Wi-Fi, and a row of dots for signal strength, allowing the eye to separate them despite sharing space. Critics have raised readability concerns, particularly in low-signal states where all three elements weaken simultaneously, making it harder to identify which connection is failing. Notably, the combined icon appears only on the cover screen; unfolding the phone restores the traditional three separate icons in iOS 27, and Control Center displays them individually as well. Apple's decision to limit the consolidated icon to a specific hardware constraint — rather than adopting it universally — has been highlighted as the more significant design choice.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Developer submits three bug-fix PRs to open-source project Rowboat in under 24 hours

A developer submitted three pull requests to Rowboat (rowboatlabs/rowboat), a 17,500-star open-source project, within roughly one day. The fixes addressed a miscategorized all-day event display in the sidebar, an unhandled server crash in child mode that now includes exponential backoff and retry limits, and a new CLI command that prints server pairing details without adding dependencies. The developer used an AI coding agent to generate code but manually reviewed every diff before submitting, catching at least one branching error that would have muddied the pull request history. During issue selection, they found multiple claimed issues with no corresponding work and two already-resolved bugs whose GitHub issues had never been closed, leaving comments to help maintainers verify. The developer concluded that reading the codebase directly — rather than filtering by beginner labels — is a more reliable way to find genuine contribution opportunities.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Cheaper AI Models May Outperform Pricier Ones Depending on Task and True Cost

A cost-performance analysis of AI models published in September 2026 highlights that a two-point capability gap between GLM-5.3-Flash and Kimi K3 masks an eightfold difference in price per task. Composite benchmark scores can mislead buyers, as individual task benchmarks reveal each model outperforms the other in different domains — GLM-5.3-Flash leads on terminal tasks while Kimi K3 leads on exam-style reasoning. Published API prices often differ from actual costs due to promotional rates, cache pricing, and mandatory reasoning tokens that cannot be disabled on some models. Kimi K3, for instance, generates 32,000 reasoning tokens per task — two-thirds of its total output — all of which are billable to the user. Processing speed also affects total infrastructure cost, with Kimi K3 taking roughly twice as long per task as GLM-5.3-Flash, a factor that can drive up compute expenses at scale.

0 comments Read more at DEV Community