Halv BlogProduct updates and engineering notes from Halv.https://halv.aiHalv used 51% fewer tokens per correct answer on SWE-rebenchhttps://halv.ai/blog/halv-swe-rebench-20-pairshttps://halv.ai/blog/halv-swe-rebench-20-pairsOn 20 paired SWE-rebench tasks, Halv used fewer tokens, cost less, and solved more tasks than vanilla Codex under the same model and verifier.Fri, 04 Sep 2026 00:00:00 GMTHow to save tokens in ChatGPT: 13 ways to cut Codex credit usagehttps://halv.ai/blog/how-to-save-tokens-in-chatgpthttps://halv.ai/blog/how-to-save-tokens-in-chatgptThirteen tested ways to cut token usage in ChatGPT and Codex: the credit rate card, model and effort choice, fast mode, caching, and AGENTS.md.Sat, 29 Aug 2026 00:00:00 GMTHow to save tokens in Claude: 14 ways to cut Claude Code usagehttps://halv.ai/blog/how-to-save-tokens-in-claudehttps://halv.ai/blog/how-to-save-tokens-in-claudeFourteen tested ways to cut token usage in Claude Code and claude.ai: context hygiene, model and effort choice, MCP pruning, hooks, and caching.Sat, 29 Aug 2026 00:00:00 GMTCodex usage limits, explained: the 5-hour window, weekly caps, and how to stretch themhttps://halv.ai/blog/codex-usage-limits-explainedhttps://halv.ai/blog/codex-usage-limits-explainedHow Codex usage limits work in 2026: the five-hour window that came back for Plus on August 25, the weekly cap every plan keeps, why Pro is exempt, what drains your allowance fastest, and how to get more from your ChatGPT plan.Thu, 27 Aug 2026 00:00:00 GMTClaude usage limits, explained: sessions, weekly caps, and how to stretch themhttps://halv.ai/blog/claude-usage-limits-explainedhttps://halv.ai/blog/claude-usage-limits-explainedHow Claude usage limits work in 2026: the 5-hour session window, weekly caps, what burns tokens fastest in Claude Code, and how to get more from your plan.Wed, 26 Aug 2026 00:00:00 GMTCrux in real sessions: 90% correct, 47% cheaper per answerhttps://halv.ai/blog/crux-sessionshttps://halv.ai/blog/crux-sessionsThe follow-up benchmark: Crux in real multi-question coding sessions on django and sympy. 90% correct answers versus grep's 58% — at 47% less cost per correct answer.Sat, 15 Aug 2026 00:00:00 GMTCrux in sessions: the benchmark that forced a result-format redesignhttps://halv.ai/blog/crux-sessions-deep-divehttps://halv.ai/blog/crux-sessions-deep-diveMethodology and full data for the session-shaped Crux benchmark, the 2.3x token blowup it exposed on sympy, and the Crux 0.6.2 output redesign that turned it into a 59% saving.Sat, 15 Aug 2026 00:00:00 GMTOne of Halv's tools made AI coding agents 45% more accuratehttps://halv.ai/blog/crux-benchmarkhttps://halv.ai/blog/crux-benchmarkWe benchmarked an AI coding agent with and without Crux, our code-index MCP server, in Codex on django and sympy: 45% more correct answers, 24% less cost per answer.Fri, 14 Aug 2026 00:00:00 GMTCrux vs grep: benchmarking an SCIP code index for AI coding agentshttps://halv.ai/blog/crux-benchmark-deep-divehttps://halv.ai/blog/crux-benchmark-deep-diveMethodology and full data for the Crux benchmark: an SCIP code-index MCP server vs grep in Codex CLI, with token-metered paired runs, ablations, and the tuning trail.Fri, 14 Aug 2026 00:00:00 GMT