ByteDance Seed 1.8

ByteDance Seed 1.8 通过 · 3 sentences

Enabling reasoning capabilities means our LLMs now run multi-step inference chains instead of generating short, direct responses, which doubles the compute resources consumed per query since each step requires processing additional context and intermediate outputs to arrive at a logical conclusion. Additionally, teams across finance, legal, and operations have rapidly adopted these reasoning tools for high-complexity workflows—like automated contract clause analysis and quarterly forecast variance checks—that were previously done manually, leading to a 20% uptick in overall query volume on top of higher per-query costs. While this has temporarily doubled our LLM bill, the time saved on labor-intensive tasks is already reducing overtime costs for senior staff, and we project these efficiency gains will offset the increased LLM expenses and generate net savings by Q4 of this year.

输出 741 tok (+581 思考) 延迟 11.1 s

ByteDance Seed 1.8 通过 · 8/8 cases

### Fixed Function ```python def longest_run(xs): if not xs: return 0 best = 1 cur = 1 for i in range(1, len(xs)): if xs[i] == xs[i-1]: cur += 1 best = max(best, cur) else: cur = 1 return best ``` ### Bug Explanation The original function had two critical bugs: 1) It initialized `best` and `cur` to 0, ignoring that any non-empty list has a minimal run length of 1 (causing single-element inputs to return 0 instead of 1). 2) It counted consecutive equal transitions (e.g., for a run of 3 elements, it counted 2 transitions) instead of the actual run length, leading to off-by-one errors for all runs—including those ending at the last index (like [2,2,2] returning 2 instead of 3). […]

输出 2072 tok (+1817 思考) 延迟 26.1 s

ByteDance Seed 1.8 通过 · 5/5 fields, guidance "withheld pending the Q4 close"

{ "company": "Meridian Robotics", "quarter": "Q3", "revenue_usd_m": 412, "growth_pct": 11, "guidance": "withheld pending the Q4 close" }

输出 757 tok (+695 思考) 延迟 10.6 s

ByteDance Seed 1.8 未命中 · 147 words, 0 banned, 1 question

Today, we’re launching a cross-provider prompt caching feature for our API gateway, built to solve key challenges for teams leveraging multiple AI service providers. Many teams reuse standardized prompts across tools but incur unnecessary latency and inflated costs from resubmitting identical requests to different platforms. This feature automatically detects and stores matching prompts across all connected providers in an encrypted, user-configurable cache, cutting redundant API calls by up to 55% and reducing average request latency by 40% for repeated prompts. It supports all major AI providers and integrates with existing gateway workflows, with options to set cache expiration times per prompt category. […]

输出 1146 tok (+970 思考) 延迟 15.2 s