DeepSeek releases V4.1-Flash-1017 with new Search and Fetch APIs

By Splitfeed EditorialLast updated Oct 17, 2026, 16:30 UTC

On October 17, 2026, DeepSeek released V4.1-Flash-1017 as an open-weight model claiming better benchmark performance, faster task completion, and shorter outputs than its predecessor. Alongside the model, DeepSeek opened its internal search engine to external developers through new Search and Fetch APIs priced at $0.75 per 1,000 calls. Users accessing the model through Claude Code and Codex receive free access during an initial promotional period. According to astrodevil1.bsky.social, the model carries an input price of $0.15 per million tokens and outperforms GPT-5.6 Sol and Opus 5 on the DeepSWE agentic coding benchmark.

The release continues a pattern DeepSeek has established of shipping models that undercut Western competitors on cost while matching or exceeding them on specific benchmarks, particularly in agentic and coding tasks. The Search and Fetch APIs are notable because they give developers access to DeepSeek's own retrieval infrastructure, not just the model itself, potentially reducing dependence on third-party search tooling. DeepSeek has not named the predecessor model being compared or specified which larger rivals the benchmarks reference, leaving room for dispute about how representative the comparisons are.

The Pro-AI camp is broadly enthusiastic. Aipulse-synestesia.bsky.social described the release as a Pareto improvement in both performance and cost, saying the numbers line up. Expertai.bsky.social highlighted coding benchmark results that it said surpass GPT and Claude in specialized tests, pointing to reinforcement learning and a large context window as key advantages for developers. Aidailypost.com framed the next question as how the model stacks against Claude Opus 5 directly. The Anti-AI camp has posted no reactions to this release; that camp would typically focus on the risks of rapidly proliferating capable open-weight models and the difficulty of auditing or controlling their downstream use. Middle Ground voices are more measured. Astrodevil1.bsky.social was direct that this is not an Opus killer but called it the first model to test for agent loops where cost matters. Chris Paxton offered a drier take, joking that using DeepSeek to verify Claude's outputs sidesteps interpretability problems only if both models cannot simultaneously be unreliable.

The clearest near-term test will be independent third-party evaluations on the DeepSWE benchmark and comparable agentic suites, along with developer reports from the free-access period through Claude Code and Codex. Those results should clarify whether the performance claims hold outside DeepSeek's own testing conditions.