{"paper":{"title":"Qwen3-Coder-Next Technical Report","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"An 80-billion-parameter model activates only three billion at inference to reach competitive results on coding agent benchmarks.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Binyuan Hui, Fan Zhou, Jiajun Zhang, Jiawei Chen, Jiaxi Yang, Jinxi Wei, Junyang Lin, Kaixin Li, Kashun Shum, Lei Zhang, Mingze Li, Mouxiang Chen, Ruisheng Cao, Wenting Zhao, Xuwu Wang, Yuheng Jing, Yunlong Feng, Zeyao Ma, Zeyu Cui, Zongmeng Zhang","submitted_at":"2026-02-28T16:25:04Z","abstract_excerpt":"We present Qwen3-Coder-Next, an open-weight language model specialized for coding agents. Qwen3-Coder-Next is an 80-billion-parameter model that activates only 3 billion parameters during inference, enabling strong coding capability with efficient inference. In this work, we explore how far strong training recipes can push the capability limits of models with small parameter footprints. To achieve this, we perform agentic training through large-scale synthesis of verifiable coding tasks paired with executable environments, allowing learning directly from environment feedback via mid-training a"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Qwen3-Coder-Next achieves competitive performance relative to its active parameter count across agent-centric benchmarks including SWE-Bench and Terminal-Bench.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That large-scale synthesis of verifiable coding tasks paired with executable environments produces training signals that generalize to real-world coding agent use cases without significant distribution shift.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"An 80B model with 3B active parameters achieves competitive coding-agent performance through agentic training on verifiable tasks and releases open weights.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"An 80-billion-parameter model activates only three billion at inference to reach competitive results on coding agent benchmarks.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"793be288c9ce18c00a6e73d3cb52a8ef94139963dc47337eddc9197d5feae86a"},"source":{"id":"2603.00729","kind":"arxiv","version":1},"verdict":{"id":"af112bd7-77f1-4b04-8d13-8fe7ecb7feb4","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T21:09:32.285141Z","strongest_claim":"Qwen3-Coder-Next achieves competitive performance relative to its active parameter count across agent-centric benchmarks including SWE-Bench and Terminal-Bench.","one_line_summary":"An 80B model with 3B active parameters achieves competitive coding-agent performance through agentic training on verifiable tasks and releases open weights.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That large-scale synthesis of verifiable coding tasks paired with executable environments produces training signals that generalize to real-world coding agent use cases without significant distribution shift.","pith_extraction_headline":"An 80-billion-parameter model activates only three billion at inference to reach competitive results on coding agent benchmarks."},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"5967965fa8969dd934ef1e66aa19bba7a09dd18d55fd01f11ded0a7db08ea862"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}