A developer published funcroute, a roughly 4,000-line C router that sits between a client and model vendors and routes each request to a specialist upstream based on its attachments, letting a cheap t…
Proteus ran a DeepSeek Harness (DSH) rc.8 self-evolution experiment in which the harness added audio input, grounded transcription, and reusable verification across 30 logical Episodes, with 27 accept…
A developer self-hosting Langfuse discovered that tracing was enabled on only one of ten agent profiles, capturing roughly 7% of traffic — 618 events across 79 traces in four weeks — while the coding …
A local model in the coder configuration running in stateful mode matched the cloud ceiling on loomcycle v1.100.0, hitting 10 out of 10 tasks, after two failed attempts to teach small local models to …
Monitoring of the deepseek/deepseek-v4-flash endpoint on open-inference/fp8 was retired with no border inputs found since 2026-09-24, and its logprob tracking stalled after recent queries failed, acco…
DeepSeek has released DeepSeek v4.1 Flash for internal beta testing, featuring a new model architecture with native multimodal support, stronger capabilities, faster speed, and lower cost. Developers …
The developer behind the zerostack CLI coding agent detailed the design of its subagent architecture, which uses parallel read-only child agents with simple text input-output to explore codebases with…
Rohit Edathil released Dagic, a typed DAG language and async execution engine that lets LLM agents compose registered tools into a typed DAG instead of calling them one at a time, avoiding the need to…
Agenthood has deployed a daily news digest agent that automatically generates release notes using an LLM, validates them against a strict contract, and opens pull requests for human review. The pipeli…
Pipe, a new programming language, outperforms Python with LangChain in five of six benchmark tasks, achieving up to 8.3× faster wall-clock time and using 40% fewer lines of code on average, according …
DeepSeek will introduce peak and off-peak pricing for its API starting Aug. 17, with off-peak rates set at half of peak rates. Peak hours are 9 a.m. to noon and 2 p.m. to 6 p.m. Beijing time. For deep…
A developer reports that the AI harness Pi from pi.dev ignored MCP server instructions and deviated from project skills during a Clojure-to-Common-Lisp port, while the Goose harness from Goose-docs.ai…
Anthropic recently removed about 80% of Claude Code's system prompt for its newest models, and a test by Ante on a small model found no performance loss. Ante A/B tested a shortened system prompt (2.3…
A test of LLM judges across 20 scenarios found that models under 4B parameters are unreliable for grading other LLMs, with qwen3:0.5b achieving only 61.5% global accuracy versus 92.0% for gemma3 and d…
A developer fabricated a claim about LLM judges failing on directional failures in a blog post, then ran a 600-judgment experiment to correct the record. The experiment found that models below 1B para…
DeepSeek's API deadline hits tomorrow, July 24 at 15:59 UTC, when the deepseek-chat and deepseek-reasoner aliases will stop working, and teams that have not migrated will face production errors. The m…
An engineer at an AI startup found that adversarial verification of AI agent outputs hits a 75% false-negative wall that standard techniques like majority voting, prompt calibration, and reruns cannot…
NaN Builders has published a comprehensive configuration guide for its LLM API, detailing setup instructions for clients, IDEs, agents, and SDKs. The guide covers available models including deepseek-v…
DeepSeek will launch its V4 series in mid-July with peak hour pricing that doubles token rates during designated periods, a first for the AI industry. The tiered pricing affects both V4-Pro and V4-Fla…
A developer warns that DeepSeek's response API is not compatible with OpenAI's Responses API, and that a common parsing mistake can silently drop the `reasoning_content` field. DeepSeek V4 models use …