{"id":"a8db8a09-f93b-495f-9b78-56e9768ba8b2","arxiv_id":"2609.37725","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Giving a language model unrestricted write access to its live context, treated as a file, beats hand-designed context-management harnesses across long-horizon agentic benchmarks at lower measured compute.","lead":"The paper makes a language model treat its own working context as an editable file it can freely rewrite, and shows existing models used this way beat fixed summarization and compaction harnesses on long-horizon coding, research, and optimization tasks, at lower measured compute. It also adds a reinforcement-learning recipe and a cache-reuse trick that cut serving cost.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"The efficiency supremacy (21.5%, 59%, 12% fewer FLOPs) rests on prefix-reuse FLOPs plus three CLM-only evaluation privileges (uncounted editing turns, up-to-six rollback retries, and a 2,048-token context-length hint that Appendix G shows materially improves the very capability driving…","rationale":"The reader's weakest-assumption call — fairness and completeness of the prefix-reuse FLOPs metric plus evaluation asymmetries favoring CLM — is the right general diagnosis, and I agree it is the load-bearing weakness. My stress-test sharpens it and shifts weight: the accuracy side of the claim is external-verifier grounded and largely robust to the metric worry, so the qualitative conclusion (CLM outperforms harnesses) is likely to survive; the fragile part is the efficiency magnitude. Among the three asymmetries, the 2,048-token hint is the most decisive because the paper's own Appendix G demonstrates that context-length estimation degrades at long contexts and that hints improve it substantially — and that capability is what efficient compaction depends on. The rollback retries are a second, independent lever because they let CLM avoid budget-overflow re-prefill costs that a fixed harness cannot. Both are admissions in the paper's own text (footnote 4; Appendix E), so my concern is not a manufactured flaw but a question of whether the headline FLOPs figures have been controlled for these affordances. The reader and I fully agree the numbers need independent re-running; I differ only in emphasis, placing the hint/rollback asymmetry ahead of the single-run variance in Table 1 and the proprietary-API baselines, which are minor contributors. I do not find an internal inconsistency that would break the central claim: Eq. 9's FLOPs accounting is self-consistent, SCR's approximation is explicitly acknowledged (b310, b317, b318), and the stepwise-GRPO training with success-gated efficiency (Eq. 6) is internally coherent even though it shares the evaluation metric. On that basis the disposition stays CONDITIONAL: the qualitative claim is sound and the magnitude claims are plausible but unverified under symmetric evaluation, so ACCEPT with conditions or a confirmed CONDITIONAL is appropriate; I keep the reader's CONDITIONAL unchanged.","tokens_in":31221,"tokens_out":3778,"duration_ms":28897,"concrete_test":"Rerun the BrowseComp-Plus comparison (Qwen3.6-27B, 32K budget, 100-turn cap) under matched conditions: (a) give the Codex-style summary baseline the same 2,048-token context-length reminder CLM receives, (b) count CLM's editing turns toward the 100-turn cap, and (c) disable CLM's rollback retries (budget overflow ends the run, as in TerminalBench). Report the accuracy gap and the prefix-reuse FLOPs gap in this fully matched setting. If the accuracy gap (59.4% vs ~53.3%) collapses below significance or the 21.5% FLOPs advantage flips, the headline efficiency and accuracy-magnitude claims overstate the mechanism's intrinsic benefit; if both hold, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim splits into accuracy supremacy and compute efficiency. The accuracy numbers are verifier-grounded (BrowseComp-Plus judge, EdgeBench judge, verifier-scored math) and are not metric-dependent, so the qualitative claim is reasonably secure. The efficiency numbers are not. They are computed on the authors' own prefix-reuse FLOPs model (Eq. 3, Appendix C), which is also the RL efficiency reward (Eq. 6) and the basis for checkpoint selection, and they are produced under conditions that favor CLM in three ways the baselines do not share: (1) CLM's editing turns do not count toward the 100-turn cap on BrowseComp-Plus (Appendix E), so CLM gets more effective decision steps at no turn-budget penalty; (2) when a request would exceed the budget, CLM rolls back the last turn and retries, up to six times (Appendix E), letting it un-commit costly context edits that would otherwise dominate re-prefill FLOPs and kill accuracy; (3) CLM receives an editing reminder 2,048 tokens before the budget (footnote 4, Appendix E), and Appendix G shows such hints substantially improve context-length estimation in the exact regime where the paper finds models degrade — and estimation of remaining budget is precisely the capability that drives efficient compaction. The paper acknowledges the hint is a 'temporary augmentation' and that outcome rewards give weak supervision for editing (b95), but it still reports the 21.5%, 59%, 12%, and 35% FLOPs figures as headline gains without reporting the sensitivity of those numbers to the hint, the retries, or the turn counting. If the 2,048-token hint alone accounts for a large share of CLM's compaction efficiency — a plausible reading of Appendix G, which shows hints 'help substantially' — then the efficiency contrast (e.g., 179 vs 437 PFLOPs on EdgeBench) partly measures an evaluation affordance rather than the context-management mechanism.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"The paper introduces Context Language Models (CLMs), which implement a model's live context as an editable file that the model can rewrite through general-purpose tools, turning context management into a native model capability rather than a harness-controlled operation. The manuscript reports that zero-shot CLMs outperform summary-based and action-constrained baselines on long-horizon agentic benchmarks: 11.4% higher accuracy with 21.5% fewer prefix-reuse FLOPs on BrowseComp-Plus, 5% higher scores with 59% fewer FLOPs on EdgeBench-10, and 65% greater downstream speedup on a six-repository agent swarm; plus gains on math optimization, an in-context skill-evolution loop on the new ContextBench, an RL phase (success-gated efficiency advantage in stepwise GRPO) improving Qwen3.5-9B from 28.8% to 42.5% on BrowseComp-Plus, and a serving-side optimization called Suffix Cache Reuse (SCR) that reuses stale KV states after in-the-middle edits.","tokens_in":31614,"tokens_out":5450,"duration_ms":39648,"significance":"If the claims hold, this is an important and timely contribution: the context-as-a-file mechanism is simple, generalizes to multi-agent and subagent settings, produces qualitative behaviors (trackers, an internal-notes role, reusable compaction functions) consistent with a real capability shift, and ships with a code release. The accuracy results are externally grounded—verifier-graded pass/fail on TerminalBench and EdgeBench and a held-out BrowseComp-Plus judge—so the qualitative claim that CLMs match or exceed state-of-the-art context-management harnesses is not circular. The fragile part is the efficiency story. The headline FLOPs figures (21.5%, 59%, 12%, 35%) are computed on the authors' own prefix-reuse FLOPs metric, which is also the RL efficiency reward, and are produced under evaluation privileges (uncounted editing turns, rollback retries, and a context-length hint) that baselines do not share. The efficiency numbers therefore need quantification before the headline claims can be taken at face value; the accuracy claims, however, are reasonably secure.","major_comments":[{"comment":"The headline 21.5% FLOPs advantage (Section 5.1.1) is computed under three CLM-only evaluation privileges described in Appendix E: CLM's editing turns do not count toward the 100-turn cap, a request that would exceed the budget is rolled back and retried up to six times, and CLM receives a context-length reminder 2,048 tokens before the budget (footnote 4). The rollback is the most damaging asymmetry: it directly lowers the re-prefill cost that prefix-reuse FLOPs measures by letting CLM un-commit context edits whose re-prefill would otherwise dominate the metric, and Appendix G shows the reminder materially improves the very context-length estimation capability that drives efficient compaction. The authors acknowledge the reminder is a 'temporary augmentation,' but the paper reports the headline efficiency numbers without any decomposition. I request ablations reporting accuracy and prefix-reuse FLOPs for CLM with (a) the hint disabled, (b) rollback retries disabled, and (c) editing turns counted against the turn cap, so the reader can see how much of the 21.5% (and the 59% EdgeBench figure) survives each privilege removed.","section":"Section 5.1.1 and Appendix E (BrowseComp-Plus setup)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The RL efficiency advantage defined in Eq. (6) rewards trajectories for minimizing the same prefix-reuse FLOPs quantity used for evaluation and for the reported cost numbers, and the RL checkpoints are selected using that metric on a validation set. The 12% efficiency gain after RL (Table 2: 1.52 to 1.34 PFLOPs per question) therefore partially reflects optimization directly against the evaluation metric rather than against an independent notion of compute. The success-gating in Eq. (6) is a reasonable guard against one class of reward hacking and the authors state the weak-supervision caveat, but what remains unresolved is whether the trained CLM is still efficient under the metric-neutral accounting once the hint-dependent budget behavior and rollback retries are removed; an ablation training with the efficiency reward but evaluating without the hinted-rollback privileges would address this.","section":"Section 4.2, Eq. (6), and Table 2","authors_acknowledge":"Outcome rewards provide only weak supervision for context editing, as successful trajectories can contain inefficient edits and failed trajectories useful ones.","authors_acknowledge_p":0.62},{"comment":"The paper reports best-of-run scores on the four math optimization problems with 'one run per method,' and the abstract's claims of up to 16.8% (Heilbronn) and 3.0% (circle packing) improvements over OpenEvolve are single-run differences with no variance estimate. Given the stochasticity of evolutionary coding-agent workflows on this timescale, a single run cannot support margins of this size; please add multiple seeds or a variance estimate, or explicitly downgrade these results to anecdotal in the abstract and main text.","section":"Table 1 and Appendix E (Math optimization)","authors_acknowledge":null,"authors_acknowledge_p":null}],"minor_comments":[{"comment":"The SCR claim of 35% compute reduction at matched performance rests on reusing stale KV cache states (Appendix B, 'the reused states were computed under the pre-edit context'), an approximation the authors flag. The Figure 9 matched-performance point should state the question count it is averaged over, since the K-sensitivity analysis in Figure 11 uses only 64 BrowseComp-Plus questions; an end-to-end accuracy check at the 830-question scale would strengthen the claim.","section":"Section 5.3 and Appendix B (SCR)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The EdgeBench-10 paragraph says 'when a request would exceed it, the harness rolls back the last turn, up to 50 times' but does not state whether this rollback privilege applies to all methods or only to CLM. Since the BrowseComp-Plus rollback is CLM-only, please specify who receives which privileges in each benchmark to make the comparisons unambiguous.","section":"Appendix E (EdgeBench-10 setup)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"Many passages have run-together words ('implementthis', 'retaininformation', 'andread–write', 'theagentis'), presumably a typesetting artifact; please fix so the text is readable.","section":"Throughout","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The ContextBench pilot uses GPT-5.4 while the main benchmarks use Qwen3.6-27B and Claude 4.6 Sonnet; a sentence stating whether the ContextBench conclusions were also checked on at least one of the main evaluation models would clarify the generality of the diagnostic.","section":"Section 3 and Appendix E (ContextBench)","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"major_revision"},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"Here's my read on the CLM paper. The core idea is genuinely new and well-executed: treating the live context as an editable file with unrestricted model write access collapses the harness-defined action space into a single primitive, and Eq. 2 formally subsuming the append-only transition is a clean way to state it. The zero-shot results on BrowseComp-Plus, TerminalBench, and the multi-agent swarm are backed by external verifiers, so the qualitative claim—models given direct context control do at least as well as human-designed harnesses—looks solid. The ContextBench diagnostic is carefully designed to isolate context management from reasoning and knowledge, and the skill-evolution and success-gated stepwise-GRPO extensions are reasonable. The SCR serving mechanism is a concrete engineering contribution, and the authors are honest that it approximates re-prefilling (b310, b317, b318).\n\nWhere I'd push back is the efficiency accounting. The stress-test note lands. The prefix-reuse FLOPs metric is the authors' own, it also serves as the RL reward, and the headline savings (21.5%, 59%, 12%, 35%) are computed under conditions that favor CLM: editing turns don't count toward the turn cap on BrowseComp-Plus, there are up to six rollback retries, and a 2,048-token context-length hint that Appendix G says substantially improves estimation. All of this is acknowledged in the paper (footnote 4, Appendix E), but the reported numbers are not conditioned on it, so the magnitude of the efficiency gains is genuinely uncertain until someone re-runs without those privileges. The math-optimization table is single-run with no variance—a real weakness, though minor in the sense that it's not the load-bearing claim. Several baselines are re-implementations and some runs use proprietary API models, which is standard for the area but limits independent verification.\n\nNone of this breaks the central claim. The accuracy results are externally grounded, and the mechanism is demonstrably more flexible than prior action-constrained tools. But the headline efficiency numbers should be read as preliminary. This deserves a serious referee: the novelty and the quality of the diagnostic work justify real reviewer time, and the efficiency claims need independent re-running before they're taken at face value. I'd cite it, and I'd bring it to reading group—the context-as-file framing is likely to be influential regardless of how the FLOPs arithmetic settles.","headline":"The core claim—models with unrestricted write access to their own context—is real and well-evidenced on accuracy, but the headline efficiency numbers rest on a self-defined metric plus CLM-only evaluation privileges, so they should be treated as conditional until re-run.","tokens_in":32370,"tokens_out":1708,"duration_ms":11730},"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek/deepseek-v4-flash","headline":"Giving a language model unrestricted write access to its own live context beats human-designed context-management harnesses on long-horizon agent tasks while using less compute.","keywords":["context language models","context management","context as a file","long-horizon agents","prefix cache reuse","reinforcement learning","skill evolution","multi-agent systems"],"falsifier":"A single controlled re-run on BrowseComp-Plus settles it: count CLM's editing turns against the 100-turn limit, give the same 2,048-token context-length hint (or none) to every method, and remove CLM's rollback retries. If CLM then fails to beat Codex-style summarization on both accuracy and prefix-reuse FLOPs, the headline gains are artifacts of the asymmetry.","tokens_in":32579,"feed_emoji":"📁","tokens_out":6821,"duration_ms":44334,"temperature":0.7,"pith_summary":"This paper argues that context management should be a native language-model capability rather than something imposed from outside. The authors implement this by mirroring the model's live context into an editable file the model can rewrite with ordinary commands, replacing the append-only transition $c_{t+1}=c_t\\oplus f^{\\mathrm{LM}}_\\theta(c_t)$ with a model-controlled $c_{t+1}=f^{\\mathrm{CLM}}_\\theta(c_t)$. On long-horizon agentic benchmarks, zero-shot CLMs built from existing models outperform summary-based and action-constrained harnesses while using fewer prefix-reuse FLOPs, and the same mechanism opens three learning routes: natural-language steering, in-context skill evolution, and online reinforcement learning. If the results hold, letting the model own and edit its context is a route to both better and cheaper long-horizon agency.","feed_headline":"Let models edit their own context and they beat fixed harnesses","feed_subtitle":"Zero-shot and RL-trained agents gained accuracy while cutting inference cost versus summary and tool-constrained managers.","key_machinery":"The mechanism is the context-as-a-file implementation: the model's live context is mirrored to a file whose path appears in the system prompt, and the model edits that file with general commands, with every edit synchronized back into the next turn's context; when the model does not edit, tokens are appended as usual. This subsumes and generalizes prior harnesses whose context-management actions are predefined. Three supporting devices carry the efficiency claims: the prefix-reuse FLOPs metric of Eq. 3, which counts prefill of the unmatched context suffix plus decode of generated tokens, thereby assigning a real cost to in-the-middle edits under prefix caching; the success-gated efficiency advantage of Eq. 6, which re-ranks only the successful trajectories within a GRPO group by inference cost so that failed trajectories never get an efficiency bonus; and Suffix Cache Reuse, which retains the cached KV states of tokens that survive an edit past the first mismatch, re-rotating their positional encodings, so that only newly inserted or appended tokens must be prefilled.","core_discovery":"The central claim is that a language model given full read-write control over its own live context can discover context-management strategies that outperform both harness-scheduled summarization and tool-constrained approaches like AutoCompact, Self-Compact, ACM, and recursive language models. Concretely, a zero-shot CLM scores 59.4% on BrowseComp-Plus with 21.5% fewer prefix-reuse FLOPs than the strongest baseline, matches Codex-style summarization on TerminalBench 2.1 with 29.5% fewer FLOPs, and exceeds it on TBLite (73.7% against 67.0%). On 12-hour EdgeBench it scores 5% higher while using 59% fewer FLOPs, and on a 24-hour six-repository swarm it achieves 65% greater downstream speedup at equal compute. The same mechanism opens three learning routes: a one-sentence instruction measurably changes the model's compaction policy; an in-context skill-evolution loop raises held-out ContextBench accuracy by up to 35.9 points at lower cost; and a success-gated efficiency reward inside stepwise GRPO lifts Qwen3.5-9B from 28.8% to 42.5% on BrowseComp-Plus, a 47.6% relative gain, while using 38.8% fewer FLOPs than a summary harness trained identically. A co-designed serving scheme, Suffix Cache Reuse, cuts server-side re-prefilling by 35% at matched performance and also helps ordinary chat serving where reasoning tokens are stripped.","pith_inferences":["If context-length awareness is the binding skill, the gains are mobile: the authors' own probe (Appendix G) shows existing models estimate their context length poorly at long lengths, so models trained with length-awareness signals could shrink or eliminate the need for the environmental hint that evaluation currently supplies.","The editable-context interface is a new prompt-injection persistence surface: a model that can rewrite its own context can carry unauthorized instructions across turns, an attack channel the paper flags and that would need defenses before broad deployment.","Because every harness context operation is expressible as a context transformation, existing harnesses act as procedural skills that could be distilled into CLM actions and then absorbed into weights, suggesting a general harness-to-CLM distillation route beyond the RL demonstrated here.","The benefit appears to scale with the model's ability to plan edits: Qwen3.5-9B edits less often and gains less than Qwen3.6-27B, which predicts that larger or more context-aware models will reap more from the CLM interface."],"forward_implications":["Zero-shot CLMs beat both harness-scheduled and action-constrained context managers across deep research, terminal coding, and single- and multi-repository optimization, at 32K budgets and, on EdgeBench, at 128K.","Context management becomes controllable in natural language: a single sentence in the prompt measurably shifts compaction threshold, semantic-boundary timing, and backup behavior with no harness or weight change.","Context-management strategies can be searched and evolved: an in-context skill-evolution loop improves held-out ContextBench accuracy by up to 35.9 points while cutting cost, and evolved skills can strictly dominate the starting point on cost-accuracy frontiers.","Context management can be internalized in model weights: stepwise GRPO with the success-gated efficiency advantage improves Qwen3.5-9B from 28.8% to 42.5% on BrowseComp-Plus while lowering inference FLOPs below a summary harness trained with the same recipe.","Suffix Cache Reuse reduces server-side re-prefilling by 35% at matched accuracy relative to standard SGLang, and the same reuse helps ordinary chat serving where reasoning tokens are stripped even when the agent never edits its context."],"supporting_citations":[{"why":"Supplies the BrowseComp-Plus deep-research benchmark, the corpus, and judging rubric used for the headline accuracy and compute comparison.","marker":"Chen et al. 2025"},{"why":"Supplies the stepwise GRPO procedure that the CLM reinforcement-learning method builds on and extends with the success-gated efficiency advantage.","marker":"Shao et al. 2024"},{"why":"Supplies the SGLang serving engine and its radix-tree prefix cache that Suffix Cache Reuse patches for post-edit suffix reuse.","marker":"Zheng et al. 2024"},{"why":"Defines the Codex CLI and its summarization prompts, the strongest baseline CLM is compared against and out-performs on BrowseComp-Plus, TerminalBench, and EdgeBench.","marker":"OpenAI 2026a"},{"why":"Provides the TerminalBench 2.1 benchmark and the Mini-SWE-Agent backbone that all methods, including CLM, share as the base harness.","marker":"Merrill et al. 2026"},{"why":"Prior art on letting a model edit a designated fixed-size memory block inside its context, which the paper positions CLM as generalizing to the whole live context.","marker":"Packer et al. 2023"},{"why":"Supplies the OpenEvolve evolutionary workflow that CLM beats on the four mathematical optimization problems.","marker":"Sharma 2025"}],"fun_headline_variants":["Let models rewrite their own context file","Self-editing context cuts compute and boosts accuracy","Language models that manage their own context","Editable context: LMs learn what to keep","Models that decide their own context, not harnesses"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The claim stands on the fairness of the prefix-reuse FLOPs metric for comparing compute and on evaluation choices that favor the CLM condition — editing turns are not counted against the 100-turn budget, CLM may roll back and retry a failed turn up to six times, and CLM receives a 2,048-token context-length hint shortly before the budget limit, an augmentation the authors concede (footnote 4) meaningfully improves length estimation; if these asymmetries bias the comparison, the headline efficiency and accuracy gains overstate the improvement.","fun_headline_variants_meta":{"raw":{"variants":["Let models rewrite their own context file","Self-editing context cuts compute and boosts accuracy","Language models that manage their own context","Editable context: LMs learn what to keep","Models that decide their own context, not harnesses"]},"model":"deepseek/deepseek-v4-flash","effort":"low","cost_usd":0.000226,"raw_usage":{"cost":0.000226268,"is_byok":false,"cost_details":{"upstream_inference_cost":0.000226268,"upstream_inference_prompt_cost":0.000131348,"upstream_inference_completions_cost":9.492e-05},"total_tokens":1482,"prompt_tokens":1143,"completion_tokens":339,"prompt_tokens_details":{"audio_tokens":0,"video_tokens":0,"cached_tokens":256,"cache_write_tokens":0},"completion_tokens_details":{"audio_tokens":0,"image_tokens":0,"reasoning_tokens":269}},"tokens_in":1143,"tokens_out":339,"cost_source":"openrouter_usage","duration_ms":3643,"image_pages":[],"temperature":1.0,"billed_attempts":1,"reasoning_tokens":269,"transport_provider":"openrouter","cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-09-30T11:23:22.972952+00:00","model_set":{"reader":"deepseek/deepseek-v4-flash"},"pipeline_version":"pith-pipeline@v1.4.1-daily-deepseek-or-medium","input_kind":null,"input_chars":null,"input_total_chars":null,"passes_missing":null,"published_standing":null,"published_confidence":null,"objection_audit":null,"run_record":null,"basis":{"kind":"not_recorded","source":null,"chars_given":null,"chars_total":null},"receipt":null,"falsifier":"A single controlled re-run on BrowseComp-Plus settles it: count CLM's editing turns against the 100-turn limit, give the same 2,048-token context-length hint (or none) to every method, and remove CLM's rollback retries. If CLM then fails to beat Codex-style summarization on both accuracy and prefix-reuse FLOPs, the headline gains are artifacts of the asymmetry.","supporting_citations":[{"cited_title":"Sglang: Efficient execution of structured language model programs","cited_arxiv_id":null,"evidence_quote":"Supplies the SGLang serving engine and its radix-tree prefix cache that Suffix Cache Reuse patches for post-edit suffix reuse."},{"cited_title":"Openevolve: an open-source evolutionary coding agent, 2025","cited_arxiv_id":null,"evidence_quote":"Supplies the OpenEvolve evolutionary workflow that CLM beats on the four mathematical optimization problems."}],"lane_accuracy":{"status":"not yet measured","lane":{"model":"deepseek/deepseek-v4-flash","input_kind":null},"url":"https://pith.science/accuracy"},"review_version":1}