{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/arxiv-2602-11683","title":"ThinkRouter: Efficient Reasoning via Routing Thinking between Latent and Discrete Spaces","arxiv_id":"2602.11683","date":"2026-02-12","proceeding":null,"authors":["Xin Xu","Tong Yu","Xiang Chen","Haoliang Wang","Julian McAuley","Saayan Mitra"],"abstract":"Recent work explores latent reasoning to improve reasoning efficiency by replacing explicit reasoning trajectories with continuous representations in a latent space, yet its effectiveness varies across settings. Analysis of model confidence dynamics under latent reasoning reveals that thinking trajectories ending in incorrect answers contain fewer low-confidence steps than those ending in correct answers. Meanwhile, we suggest that soft embeddings aggregated by multiple low-confidence thinking alternatives may introduce and propagate noise, leading to high confidence in unreliable reasoning trajectories. Motivated by these observations, ThinkRouter, an inference-time confidence-aware routing mechanism is proposed to avoid high confidence and noise for efficient reasoning. ThinkRouter routes thinking to the discrete token space when model confidence is low, and to the latent space otherwise. Extensive experiments on STEM reasoning and coding benchmarks across diverse large reasoning models demonstrate that ThinkRouter outperforms explicit CoT, random routing, and latent reasoning baselines in terms of accuracy, achieving an average improvement of 19.70 points in Pass@1, while reducing generation length by up to 15.55%. Further comprehensive analysis reveals that ThinkRouter can calibrate errors arising from explicit CoT and latent reasoning, and accelerates end-of-thinking token generation by globally lowering model confidence.","url_abs":"https://arxiv.org/abs/2602.11683","url_pdf":"https://arxiv.org/pdf/2602.11683","source":{"archive":null,"snapshot":"2025-07-28","note":"not in the Papers with Code archive (frozen at the snapshot)","row_kind":"graph","title_abstract_authors_date":"arXiv metadata, CC0 1.0 (https://info.arxiv.org/help/license)"},"code_links":[],"tasks":[],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2602.11683","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2602.11683"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"mentioned_in_github":null,"is_official":null,"provenance":"deterministic:regex_extraction","mentioned_in_paper":null,"url":"https://github.com/eric-ai-lab/Soft-Thinking","reach":{"status":"ok","spdx":"MIT"}}],"summary":{"ran":8,"unverified":8},"by_repo_kind":{"found_in_text":{"samples":16,"ran":8,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"64f004b9216b7322","entry":"benchmark_sequential_vs_batch","repo":"eric-ai-lab/Soft-Thinking","repo_kind":"found_in_text","path":"sglang_soft_thinking_pkg/benchmark/benchmark_batch/benchmark_tokenizer.py","file_url":"https://github.com/eric-ai-lab/Soft-Thinking/blob/HEAD/sglang_soft_thinking_pkg/benchmark/benchmark_batch/benchmark_tokenizer.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"64f004b9216b7322"}},{"code_sha256_prefix":"83900d0c3ab83065","entry":"generate_event_triple_prompt","repo":"eric-ai-lab/Soft-Thinking","repo_kind":"found_in_text","path":"sglang_soft_thinking_pkg/benchmark/generative_agents/agent_functions.py","file_url":"https://github.com/eric-ai-lab/Soft-Thinking/blob/HEAD/sglang_soft_thinking_pkg/benchmark/generative_agents/agent_functions.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"83900d0c3ab83065"}},{"code_sha256_prefix":"577ff88cf5082d20","entry":"generate_pronunciatio_prompt","repo":"eric-ai-lab/Soft-Thinking","repo_kind":"found_in_text","path":"sglang_soft_thinking_pkg/benchmark/generative_agents/agent_functions.py","file_url":"https://github.com/eric-ai-lab/Soft-Thinking/blob/HEAD/sglang_soft_thinking_pkg/benchmark/generative_agents/agent_functions.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"577ff88cf5082d20"}},{"code_sha256_prefix":"9afb749e3f5b90cf","entry":"generate_random_prompt","repo":"eric-ai-lab/Soft-Thinking","repo_kind":"found_in_text","path":"sglang_soft_thinking_pkg/benchmark/benchmark_batch/benchmark_batch.py","file_url":"https://github.com/eric-ai-lab/Soft-Thinking/blob/HEAD/sglang_soft_thinking_pkg/benchmark/benchmark_batch/benchmark_batch.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"9afb749e3f5b90cf"}},{"code_sha256_prefix":"5139455c0f07d032","entry":"generate_random_prompts","repo":"eric-ai-lab/Soft-Thinking","repo_kind":"found_in_text","path":"sglang_soft_thinking_pkg/benchmark/benchmark_batch/benchmark_tokenizer.py","file_url":"https://github.com/eric-ai-lab/Soft-Thinking/blob/HEAD/sglang_soft_thinking_pkg/benchmark/benchmark_batch/benchmark_tokenizer.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"5139455c0f07d032"}},{"code_sha256_prefix":"19faa93c902e1c91","entry":"generate_unique_prefix","repo":"eric-ai-lab/Soft-Thinking","repo_kind":"found_in_text","path":"sglang_soft_thinking_pkg/benchmark/bench_in_batch_prefix/bench_in_batch_prefix.py","file_url":"https://github.com/eric-ai-lab/Soft-Thinking/blob/HEAD/sglang_soft_thinking_pkg/benchmark/bench_in_batch_prefix/bench_in_batch_prefix.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"19faa93c902e1c91"}},{"code_sha256_prefix":"e4b32a865322152c","entry":"poignancy_event_prompt","repo":"eric-ai-lab/Soft-Thinking","repo_kind":"found_in_text","path":"sglang_soft_thinking_pkg/benchmark/generative_agents/agent_functions.py","file_url":"https://github.com/eric-ai-lab/Soft-Thinking/blob/HEAD/sglang_soft_thinking_pkg/benchmark/generative_agents/agent_functions.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"e4b32a865322152c"}},{"code_sha256_prefix":"5bc9a6f07218ab7b","entry":"send_batch_request","repo":"eric-ai-lab/Soft-Thinking","repo_kind":"found_in_text","path":"sglang_soft_thinking_pkg/benchmark/benchmark_batch/benchmark_batch.py","file_url":"https://github.com/eric-ai-lab/Soft-Thinking/blob/HEAD/sglang_soft_thinking_pkg/benchmark/benchmark_batch/benchmark_batch.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"5bc9a6f07218ab7b"}},{"code_sha256_prefix":"70f1d10961d52f6d","entry":"generate_random_string","repo":"eric-ai-lab/Soft-Thinking","repo_kind":"found_in_text","path":"sglang_soft_thinking_pkg/benchmark/bench_in_batch_prefix/bench_in_batch_prefix.py","file_url":"https://github.com/eric-ai-lab/Soft-Thinking/blob/HEAD/sglang_soft_thinking_pkg/benchmark/bench_in_batch_prefix/bench_in_batch_prefix.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"70f1d10961d52f6d"}},{"code_sha256_prefix":"4c13c3bdb761b46c","entry":"get_answer_value","repo":"eric-ai-lab/Soft-Thinking","repo_kind":"found_in_text","path":"sglang_soft_thinking_pkg/benchmark/gsm8k/bench_other.py","file_url":"https://github.com/eric-ai-lab/Soft-Thinking/blob/HEAD/sglang_soft_thinking_pkg/benchmark/gsm8k/bench_other.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"4c13c3bdb761b46c"}},{"code_sha256_prefix":"9e0d3df1a2d9a123","entry":"get_few_shot_examples","repo":"eric-ai-lab/Soft-Thinking","repo_kind":"found_in_text","path":"sglang_soft_thinking_pkg/benchmark/gsm8k/bench_other.py","file_url":"https://github.com/eric-ai-lab/Soft-Thinking/blob/HEAD/sglang_soft_thinking_pkg/benchmark/gsm8k/bench_other.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"9e0d3df1a2d9a123"}},{"code_sha256_prefix":"06681001174b0ca9","entry":"get_few_shot_examples","repo":"eric-ai-lab/Soft-Thinking","repo_kind":"found_in_text","path":"sglang_soft_thinking_pkg/benchmark/hellaswag/bench_other.py","file_url":"https://github.com/eric-ai-lab/Soft-Thinking/blob/HEAD/sglang_soft_thinking_pkg/benchmark/hellaswag/bench_other.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"06681001174b0ca9"}},{"code_sha256_prefix":"2e994f473fac61f9","entry":"get_one_example","repo":"eric-ai-lab/Soft-Thinking","repo_kind":"found_in_text","path":"sglang_soft_thinking_pkg/benchmark/gsm8k/bench_other.py","file_url":"https://github.com/eric-ai-lab/Soft-Thinking/blob/HEAD/sglang_soft_thinking_pkg/benchmark/gsm8k/bench_other.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"2e994f473fac61f9"}},{"code_sha256_prefix":"5b4a33e73835b337","entry":"get_one_example","repo":"eric-ai-lab/Soft-Thinking","repo_kind":"found_in_text","path":"sglang_soft_thinking_pkg/benchmark/hellaswag/bench_other.py","file_url":"https://github.com/eric-ai-lab/Soft-Thinking/blob/HEAD/sglang_soft_thinking_pkg/benchmark/hellaswag/bench_other.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"5b4a33e73835b337"}},{"code_sha256_prefix":"60d747ff8e6ab539","entry":"prepare_all_prompts","repo":"eric-ai-lab/Soft-Thinking","repo_kind":"found_in_text","path":"sglang_soft_thinking_pkg/benchmark/benchmark_batch/benchmark_batch.py","file_url":"https://github.com/eric-ai-lab/Soft-Thinking/blob/HEAD/sglang_soft_thinking_pkg/benchmark/benchmark_batch/benchmark_batch.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"60d747ff8e6ab539"}},{"code_sha256_prefix":"e5914c7114b177c6","entry":"prepare_prompts","repo":"eric-ai-lab/Soft-Thinking","repo_kind":"found_in_text","path":"sglang_soft_thinking_pkg/benchmark/bench_in_batch_prefix/bench_in_batch_prefix.py","file_url":"https://github.com/eric-ai-lab/Soft-Thinking/blob/HEAD/sglang_soft_thinking_pkg/benchmark/bench_in_batch_prefix/bench_in_batch_prefix.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"e5914c7114b177c6"}}]},"arxiv_metadata":{"licence":"arXiv metadata, CC0 1.0 (https://info.arxiv.org/help/license)","fields":["title","abstract","authors","date"],"primary_category":"cs.LG","source":"arxiv_2026.jsonl"},"syntology_extracted_results":null}