The case concerns a hybrid coding-agent architecture in which a frontier model handles planning or orchestration while cheaper models execute delegated work, aiming to retain quality while reducing inference costs. The supplied snippets support growing interest in such hybrid workflows, rising coding-agent costs, and a narrowing performance gap between frontier and cheaper or self-hostable models. However, they do not substantiate the named “Fable 5” system, Anthropic’s alleged benchmark, the specific 96%-performance/46%-cost figures, or the claim of independent confirmation.
2026-07-24T11:23:22Z
The broad half-cost, near-frontier-performance hypothesis has been superseded by a topology-specific conclusion: selective gating can save materially, while delegated fan-out and context handoffs can increase cost and degrade reliability. Further adoption anecdotes will not resolve this without controlled coding-agent evaluations by routing strategy.
2026-07-24T10:25:45Z
No legible new result changes the emerging conclusion: multi-model orchestration is spreading, but cost savings depend on selective gating, task granularity, and harness design, while delegation fan-out can increase cost. The original near-frontier-performance-at-half-cost claim remains unconfirmed and should not be generalized from adoption alone.
2026-07-24T09:22:34Z
The trigger adds no substantive evidence beyond the already absorbed selective-gating example. Adoption remains broad, but the economics have resolved into a topology- and harness-dependent question rather than general support for near-frontier coding performance at roughly half cost.
2026-07-24T08:23:35Z
The cheap-classifier implementation strengthens selective gating as the more credible cost-saving topology: avoid unnecessary frontier calls rather than fan work out to cheaper agents. It reinforces that savings are harness-dependent but adds no coding-quality comparison, leaving the near-frontier-performance-at-half-cost claim unconfirmed.
2026-07-24T08:21:10Z
evidence attached: reddit.post.1v54psu — Using a cheap classifier to route only substantive requests to an expensive model is a concrete cost-aware model-orchestration pattern.
2026-07-24T07:26:11Z
The latest trigger adds no substantive evidence beyond the already absorbed reports of delegation overhead and context loss. Adoption continues to spread, but savings now appear strongly topology- and harness-dependent, with the coding-specific near-frontier-performance-at-half-cost claim still unconfirmed.
2026-07-24T06:23:33Z
The new hands-on report strengthens the countercase that delegation can increase token use and damage context continuity, making savings highly dependent on routing topology, task granularity, and harness design. Broad implementation is established, but the claimed near-frontier coding performance at roughly half cost now looks conditional rather than generally supported.
2026-07-24T06:20:59Z
evidence attached: reddit.post.1v52kfp — User experience highlights substantial token-cost and context-window failure modes when frontier agents delegate to worker models.
2026-07-24T05:22:20Z
No legible new evidence or engagement movement advances the quality-adjusted cost claim; this is repetitive amplification of an already widespread implementation pattern. Adoption remains strong, but topology-dependent overhead and the coding-specific near-frontier-performance-at-half-cost hypothesis remain unresolved.
2026-07-24T04:22:15Z
The trigger exposes no identifiable new evidence beyond implementations and results already absorbed, so it is repetitive amplification rather than further confirmation. The architecture is spreading, but topology-dependent overhead and the coding-specific near-frontier-performance-at-half-cost claim remain unresolved.
2026-07-24T03:28:56Z
The trigger exposes no legible new quality-adjusted cost result and is further repetitive amplification. Multi-model coding orchestration is broadly implemented, but topology-dependent overhead and the coding-specific near-frontier-performance-at-half-cost claim remain unresolved.
2026-07-24T02:21:58Z
The apparent attachment adds no discernible independent result beyond evidence already absorbed, so it is further repetitive amplification. Adoption is broad, but topology-dependent overhead and the coding-specific near-frontier-performance-at-half-cost claim remain unresolved.
2026-07-24T01:26:41Z
The apparent update adds no legible independent measurement beyond already absorbed implementations, so it is repetitive amplification. Adoption is still spreading, but topology-dependent overhead and the coding-specific near-frontier-performance-at-half-cost claim remain unresolved.
2026-07-24T00:20:53Z
The trigger exposes no identifiable new result beyond evidence already absorbed, so it is further repetitive amplification. Adoption remains broad, but topology-dependent overhead and the coding-specific near-frontier-performance-at-half-cost claim remain unresolved.
2026-07-23T23:26:30Z
The latest activity is repetitive amplification of evidence already absorbed, not a new controlled coding-agent quality-versus-cost result. Adoption remains broad, but topology-dependent overhead and the near-frontier-performance-at-half-cost claim remain unresolved.
2026-07-23T22:28:10Z
The refreshed discussion adds only modest engagement to evidence already absorbed, with no new controlled coding-agent quality-versus-cost result. Adoption remains broad and accelerating, but topology-dependent overhead and the near-frontier-performance-at-half-cost claim remain unresolved.
2026-07-23T21:27:16Z
The latest router adds another independent implementation but no quality-adjusted cost measurement, so it reinforces adoption without advancing the quantitative hypothesis. The pattern is spreading, while topology-dependent overhead and the near-frontier-performance-at-half-cost claim remain unresolved.
2026-07-23T21:21:19Z
evidence attached: hn.story.49027640 — A small independent implementation of routing Claude Code through alternative models materially contextualizes emerging multi-model agent orchestration, though it does not validate performance or savings.
2026-07-23T20:25:01Z
Echo strengthens the case that selective routing across cheaper models can approach frontier quality, and underscores that economics depend on routing topology rather than orchestration alone. Its results remain self-reported and not coding-agent-specific, so they do not independently confirm near-frontier coding performance at roughly half cost.
2026-07-23T20:21:16Z
evidence attached: hn.story.49026810 — A new implementation provides direct evidence for the open case that combining specialized open models can approach frontier quality at lower cost.
2026-07-23T18:27:08Z
No legible new evidence extends the already absorbed Subagent Tax result; the update is repetitive activity rather than a fresh quality-versus-cost measurement. The orchestration pattern remains widespread, but topology-dependent overhead and the specific near-frontier-performance-at-half-cost claim remain unresolved.
2026-07-23T16:23:34Z
The Subagent Tax is the first substantive counterevidence: an independent measurement showing subagent fan-out can cost 2-6x more tokens with no speed gain, directly challenging the orchestration-saves-cost premise rather than just adding another implementation anecdote. This doesn't disprove the frontier-orchestrator/cheap-worker pattern specifically (different topology — parallel fan-out vs. sequential delegation) but it complicates the general 'multi-model coordination cuts cost' narrative and warrants tracking as genuine counterevidence going forward. The core 96%-performance/46%-cost claim remains unconfirmed either way.
2026-07-23T16:21:25Z
evidence attached: hn.story.49023564 — This is independent evidence that subagent fan-out can impose large token and latency costs, potentially contradicting the presumed orchestration savings.
2026-07-23T15:23:00Z
The apparent attachment provides no legible independent quality-versus-cost measurement and is further amplification of an established, spreading implementation pattern. Practical orchestration is well supported, but the coding-specific claim of near-frontier performance at roughly half the cost remains unconfirmed.
2026-07-23T14:23:42Z
The latest trigger adds no legible independent measurement and is further repetitive amplification of a well-established, spreading implementation pattern. The coding-specific claim of near-frontier performance at roughly half the cost remains unconfirmed.
2026-07-23T13:33:58Z
The latest attachment adds no discernible independent quality-versus-cost result, extending repetitive amplification of an established and spreading implementation pattern. The architecture is practical, but the coding-specific claim of near-frontier performance at roughly half the cost remains unconfirmed.
2026-07-23T12:26:45Z
The latest trigger adds no identifiable independent result beyond implementations already absorbed, making it further repetitive amplification. Multi-model coding orchestration is established and spreading, but controlled confirmation of near-frontier coding performance at roughly half the cost remains absent.
2026-07-23T11:22:06Z
The trigger exposes no identifiable new result beyond evidence already absorbed, so it is further repetitive amplification. Multi-model coding orchestration is established and spreading, but independent controlled confirmation of near-frontier coding performance at roughly half the cost remains absent.
2026-07-23T10:31:46Z
The trigger reveals no legible new result beyond evidence already absorbed, so it is further repetitive amplification. Multi-model coding orchestration is established and spreading, but independent controlled confirmation of near-frontier coding performance at roughly half the cost remains absent.
2026-07-23T09:21:35Z
The apparent new attachment adds no legible independent quality-versus-cost measurement and is further amplification of an already established, spreading implementation pattern. The coding-specific claim of near-frontier performance at roughly half the cost remains unconfirmed.
2026-07-23T08:22:09Z
The trigger exposes no legible new result beyond implementations and engagement already absorbed, so it does not change the case’s meaning. Multi-model coding orchestration is established and spreading, but independent controlled confirmation of near-frontier coding performance at roughly half the cost remains absent.
2026-07-23T07:22:51Z
The apparent update adds no legible independent measurement beyond implementations already absorbed, so it is repetitive amplification. Multi-model coding orchestration is established and spreading, but the coding-specific claim of near-frontier performance at roughly half the cost remains unconfirmed.
2026-07-23T06:26:56Z
The latest attachment and reobservations add another implementation but no independent controlled quality-versus-cost result. Multi-model coding orchestration is established and spreading, while the specific near-frontier-performance-at-half-cost claim remains unconfirmed.
2026-07-23T05:21:55Z
Alloy’d adds another packaged, working router and reinforces that cross-provider coding-agent orchestration is spreading as an engineering pattern. It optimizes subscription capacity rather than measuring quality-adjusted inference cost, so it does not advance the specific near-frontier-performance-at-half-cost claim.
2026-07-23T05:20:50Z
evidence attached: reddit.post.1v43tav — A working Claude Code/Codex router is direct implementation evidence that users are orchestrating multiple model providers to manage cost and capacity.
2026-07-23T04:21:54Z
The trigger adds no identifiable substantive evidence beyond implementations and engagement already absorbed. Multi-model coding orchestration is spreading as an engineering pattern, but independent controlled confirmation of near-frontier coding performance at roughly half the cost remains absent.
2026-07-23T03:22:05Z
The trigger exposes no substantive new evidence beyond already absorbed implementations and engagement. Multi-model coding orchestration is clearly spreading, but independent controlled confirmation of near-frontier coding performance at roughly half the cost remains absent.
2026-07-23T02:24:25Z
The apparent update contains no legible new evidence beyond reobservations and engagement, so it is repetitive amplification. Independent implementations show the orchestration pattern is spreading, but controlled coding-agent evidence for near-frontier performance at roughly half cost remains absent.
2026-07-23T01:21:32Z
The latest trigger exposes no substantive new evidence beyond reobservations of already absorbed implementations. Multi-model coding orchestration is spreading, but independent controlled confirmation of near-frontier performance at roughly half the cost remains absent.
2026-07-23T00:21:04Z
The latest trigger is reobservation and engagement around Echo rather than new independent evidence. Multi-model routing continues to spread, but the coding-specific claim of near-frontier performance at roughly half cost remains unconfirmed.
2026-07-22T23:22:49Z
Echo adds a directly aligned adaptive-routing product claim, suggesting the target economics are attracting concrete implementations beyond planner–worker harnesses. Its results are internal, lightly observed, and not specific to coding agents, so it does not independently confirm near-frontier coding performance at roughly half cost.
2026-07-22T23:20:57Z
evidence attached: hn.story.49011856 — This is a direct product claim that adaptive routing across open models can approach frontier quality at roughly half or less of the inference cost.
2026-07-22T22:23:17Z
The trigger exposes no legible new result beyond reobservations of evidence already absorbed. Independent implementations establish a spreading orchestration pattern, but the specific near-frontier coding performance at roughly half cost remains unconfirmed.
2026-07-22T21:23:10Z
The apparent update is repetitive amplification of implementations already absorbed, not a new controlled coding quality-versus-cost result. The routing pattern remains established and spreading, while the specific near-frontier performance at roughly half cost remains unconfirmed.
2026-07-22T20:31:35Z
The latest trigger exposes no substantive new result beyond evidence already absorbed. Multi-model coding orchestration is clearly spreading through independent implementations, but controlled confirmation of near-frontier coding performance at roughly half the cost remains absent.
2026-07-22T19:30:02Z
The latest trigger adds no substantive evidence beyond already absorbed implementations and engagement. Multi-model routing is spreading as an engineering pattern, but controlled confirmation of near-frontier coding performance at roughly half the cost remains absent.
2026-07-22T18:36:25Z
Confidence-based local-to-cloud escalation adds a distinct routing mechanism and strengthens evidence that cheaper-worker orchestration is becoming implementable rather than merely anecdotal. It still provides no controlled coding-agent quality-versus-cost comparison, leaving the roughly half-cost, near-frontier-performance claim unresolved.
2026-07-22T18:22:05Z
evidence attached: hn.story.49011101 — A roster-based coding-agent work-board orchestrator is relevant evidence for emerging multi-agent coordination patterns, though the low-signal listing provides no performance data.
2026-07-22T18:22:05Z
evidence attached: hn.story.49010782 — This independently supports the broader cost-saving model-routing pattern by using local Gemma confidence to escalate only a minority of queries to a stronger cloud model.
2026-07-22T18:22:05Z
evidence attached: reddit.post.1v3nw3j — This provides a concrete local-to-cloud confidence-routing implementation that materially supports the case's cheaper-worker orchestration hypothesis.
2026-07-22T17:27:58Z
The new publishing deployment broadens evidence that orchestration-first workflows are practical, but it is outside coding and supplies no comparative cost or quality measurements. The architecture continues spreading, while handoff reliability and the specific near-frontier-performance-at-half-cost claim remain unresolved.
2026-07-22T17:21:44Z
evidence attached: hn.story.49009663 — A large real-world orchestration deployment bears on whether multi-agent, multi-model workflows are becoming a practical pattern, though it does not yet validate the cost claim.
2026-07-22T15:28:41Z
The latest trigger exposes no legible new evidence beyond material already absorbed, so it is repetitive amplification. Multi-model coding orchestration is clearly spreading through implementations, but handoff reliability and the specific near-frontier-performance-at-half-cost claim remain unresolved.
2026-07-22T14:29:00Z
The trigger exposes no identifiable new result beyond evidence already absorbed, so it is repetitive amplification rather than further confirmation. The orchestration pattern is spreading, but handoff reliability and controlled evidence for near-frontier coding performance at roughly half the cost remain unresolved.
2026-07-22T13:30:06Z
The new hands-on report introduces a modest reliability counterweight: planner–worker delegation can stall at the handoff, making harness quality and recovery behavior central to the economics. It does not materially advance the unresolved claim of near-frontier coding performance at roughly half the cost.
2026-07-22T13:21:50Z
evidence attached: reddit.post.1v3eqwz — This is an additional user report of using a frontier model as planner and a cheaper model as worker, though the reported stalling limits its evidentiary strength.
2026-07-22T12:24:35Z
The trigger exposes no identifiable new result beyond evidence already absorbed. Packaged implementations and independent experiments show the orchestration pattern is spreading, but controlled confirmation of near-frontier coding performance at roughly half the cost remains absent.
2026-07-22T11:25:37Z
The latest trigger adds no substantive result beyond the packaged implementations already absorbed. The orchestration pattern is spreading, but independent controlled confirmation of near-frontier coding performance at roughly half the cost remains absent.
2026-07-22T10:28:22Z
Frugal adds a concrete, independently built Claude Code router with objective escalation checks, strengthening the shift from ad hoc delegation toward packaged implementations. It reports no comparative outcomes, so the claim of near-frontier coding performance at roughly half the cost remains unconfirmed.
2026-07-22T10:20:59Z
evidence attached: reddit.post.1v3b4pt — Frugal is a concrete Claude Code implementation of routing mechanical subtasks to cheaper workers and escalating on objective checks, directly testing the open orchestration hypothesis.
2026-07-21T23:29:25Z
The trigger adds no identifiable evidence beyond already absorbed implementations and cost-routing claims. The architecture continues spreading, but independent controlled confirmation of near-frontier coding performance at roughly half the cost remains absent.
2026-07-21T22:23:10Z
The best-execution item adds another cost-routing implementation claim, but its title-level evidence does not establish coding-agent quality retention or independently validate the roughly 50% savings target. The broader orchestration pattern continues spreading, while the case’s core quantitative hypothesis remains unsettled.
2026-07-21T22:21:11Z
evidence attached: hn.story.48998844 — The best-execution approach directly bears on whether model routing can cut LLM costs while preserving task-level intelligence.
2026-07-21T21:26:13Z
The apparent attachment adds no identifiable evidence beyond material already absorbed, so this is repetitive amplification rather than a new result. Multi-model coding orchestration is spreading and being empirically refined, but independent confirmation of near-frontier coding performance at roughly half the cost remains absent.
2026-07-21T19:25:57Z
The trigger contains no identifiable new evidence beyond material already absorbed, so the case’s meaning is unchanged. Multi-model coding orchestration is spreading and undergoing empirical refinement, but independent confirmation of near-frontier performance at roughly half the cost remains absent.
2026-07-21T18:26:57Z
The latest movement is discussion growth around already absorbed evidence, not a new independent quality-versus-cost result. Implementations and empirical refinement keep the pattern accelerating, but the specific near-frontier performance at roughly half cost remains unconfirmed.
2026-07-21T17:38:08Z
The follow-up benchmark adds executed, independently designed evaluation rather than another implementation anecdote, showing that cheap-worker viability depends on task knowledge and routing rather than reasoning alone. This advances the pattern into active empirical refinement, though the specific near-frontier performance at roughly half cost is still not independently established.
2026-07-21T17:21:53Z
evidence attached: reddit.post.1v2nvyc — Independent multi-model coding-agent evaluation with executed verification materially supports the case's cost-performance hypothesis.
2026-07-21T16:32:36Z
The routing proxy, adversarial review harness, and token-cost report broaden evidence that multi-model coding workflows are spreading into distinct implementations. None provides a controlled quality-versus-cost comparison, so the specific near-frontier performance at roughly half cost remains unconfirmed.
2026-07-21T16:22:01Z
evidence attached: hn.story.48993565 — Independent reporting on coding-assistant strategies that reduce token bills bears directly on whether cheaper-model orchestration delivers substantial cost savings.
2026-07-21T16:22:01Z
evidence attached: hn.story.48993960 — An adversarial Claude/GPT coding-review harness provides practical evidence about multi-model orchestration in coding workflows.
2026-07-21T16:22:01Z
evidence attached: hn.story.48993540 — A self-training model-routing proxy is directly relevant implementation evidence for whether routing across models can reduce inference cost.
2026-07-21T15:31:09Z
The latest trigger adds no substantive evidence beyond unchanged engagement and previously absorbed implementations. Multi-model coding orchestration is established as practical, but independent controlled confirmation of near-frontier performance at roughly half the cost remains absent.
2026-07-21T14:29:09Z
The apparent attachment adds no identifiable independent result and is further repetitive amplification. The orchestration pattern is established, but the specific near-frontier coding quality at roughly half cost remains unverified.
2026-07-21T13:22:47Z
The trigger adds no identifiable independent measurement beyond evidence already absorbed. The orchestration pattern is established as practical, but the specific near-frontier coding performance at roughly half cost remains unverified.
2026-07-21T12:22:33Z
The latest trigger is repetitive amplification of evidence already absorbed, not a new quality-versus-cost result. Multi-model coding orchestration is established as practical, but independent confirmation of near-frontier performance at roughly half the cost remains absent.
2026-07-21T11:26:28Z
The newly attached item duplicates an already absorbed discussion and adds no controlled coding quality-versus-cost result. The orchestration pattern is established, but the specific near-frontier performance at roughly half cost remains unverified.
2026-07-21T11:20:59Z
evidence attached: reddit.post.1v2fix0 — shared external link with case evidence
2026-07-21T10:22:39Z
The new trigger is only reobservation and engagement around evidence already absorbed, not an independent quality-versus-cost result. The architecture is established as practical, but the specific near-frontier performance at roughly half cost remains unverified.
2026-07-21T09:22:56Z
The trigger is another reobservation with no new controlled quality-versus-cost evidence. Independent implementations establish the orchestration pattern, but the claimed near-frontier coding performance at roughly half cost remains unverified.
2026-07-21T08:22:53Z
The latest trigger contains no identifiable new evidence beyond reobservations, so it does not change the case. Multi-model coding orchestration is established as practical, but controlled confirmation of near-frontier performance at roughly half the cost remains absent.
2026-07-21T07:23:05Z
The latest attachments are reobservations and engagement around evidence already absorbed, not new independent measurements. The orchestration pattern is established as practical, but the specific near-frontier coding performance at roughly half cost remains unverified.
2026-07-21T06:27:39Z
The apparent new attachment is only reobservation of evidence already absorbed, so it adds no independent measurement of coding quality versus cost. Multi-model orchestration is established as practical, but the specific near-frontier performance at roughly half cost remains unverified.
2026-07-21T05:29:48Z
The apparent update is reobservation and engagement around evidence already absorbed, not a new independent measurement. The orchestration pattern is established as practical, but near-frontier coding performance at roughly half the cost remains unconfirmed.
2026-07-21T04:22:14Z
The latest activity adds no identifiable substantive evidence beyond repeated implementation anecdotes. The architecture is established as practical, but controlled confirmation of near-frontier coding quality at roughly half the cost remains absent.
2026-07-21T03:26:15Z
The latest hands-on report adds another weak implementation example but no measured quality or cost comparison. Multi-model coding orchestration remains well corroborated as a practical pattern, while the near-frontier performance at roughly half cost claim is still unverified.
2026-07-21T03:21:07Z
evidence attached: reddit.post.1v26kk6 — Independent hands-on use supports the hypothesis that a stronger model can delegate focused work to a cheaper model with possible usage savings.
2026-07-21T02:22:12Z
The latest activity is repetitive amplification rather than new controlled evidence. Multi-model coding orchestration is established as a practical pattern, but the claimed near-frontier performance at roughly half the cost remains unverified.
2026-07-21T01:25:38Z
The new attachment appears to be another reobservation rather than independent controlled evidence. The implementation pattern is well corroborated, but the specific near-frontier coding performance at roughly half cost remains unverified.
2026-07-21T00:22:25Z
The latest reobservations add no substantive evidence beyond the already-corroborated implementation pattern. Practicality is established, but controlled evidence for near-frontier coding performance at roughly half the cost remains absent.
2026-07-20T23:23:54Z
The latest activity is amplification of an already-corroborated implementation pattern, not new controlled evidence on coding quality versus cost. The architecture remains practical and economically plausible, while the specific 96%-performance-at-46%-cost hypothesis remains unverified.
2026-07-20T22:24:50Z
Selective frontier-model use adds another independent implementation pattern, reinforcing that cheap-worker routing is practical rather than merely proposed. It still supplies no controlled coding-quality or cost comparison, so the specific near-frontier performance at roughly half cost claim remains unverified.
2026-07-20T22:21:07Z
evidence attached: hn.story.48916512 — The selective use of a frontier model for only one coding edit supports the case that cheaper workers can handle most agent work.
2026-07-20T21:25:11Z
Ramp’s production router adds consequential evidence that multi-model routing can deliver material cost savings, strengthening the economic premise beyond hobbyist implementations. It still does not measure coding-agent quality or validate the claimed near-half cost at roughly preserved performance, so the core quantitative hypothesis remains unsettled.
2026-07-20T21:21:05Z
evidence attached: hn.story.48984923 — A real production model router reporting a 30% internal cost reduction materially supports the case for orchestrated multi-model inference.
2026-07-20T20:28:56Z
The new model-economics discussion reinforces that cheaper-worker orchestration is becoming a recognized implementation pattern, but adds no independent quality-versus-cost measurements. The architecture is corroborated; the specific 96%-performance-at-46%-cost claim remains unverified.
2026-07-20T20:21:19Z
evidence attached: hn.story.48982535 — Cursor's discussion of agent swarms and model economics materially contextualizes whether cheaper worker models can make orchestration economical.
2026-07-20T19:25:15Z
Multiple independent implementations, including a 198-run benchmark, now corroborate that frontier-led delegation to cheaper or local coding models is practical. They still do not independently confirm the headline claim of retaining roughly 96% of performance at about half the cost, so the routing economics remain unsettled.
2026-07-20T19:21:11Z
evidence attached: reddit.post.1v1tnmn — Independent 198-run testing provides useful corroboration that MCP-based delegation can combine frontier, low-cost, and local worker models within a coding-agent workflow.
2026-07-20T19:21:11Z
evidence attached: reddit.post.1v1tk6d — Direct independent use of a frontier orchestrator delegating work to a local model bears on whether worker-agent setups are reliable and cost-effective.
2026-07-20T19:21:11Z
evidence attached: reddit.post.1v1um8l — Anecdotal evidence that hosted frontier-agent products may displace extensible local agent platforms, contextualising the orchestration-versus-integration tradeoff.
2026-07-20T17:30:15Z
The standalone harness adds independent evidence that two-model coding orchestration is being implemented, but it provides no comparative quality or cost measurements. The architecture is increasingly familiar; the specific claim of preserving roughly 96% of performance at about half the cost remains unconfirmed.
2026-07-20T17:21:29Z
evidence attached: hn.story.48981521 — A standalone two-model coding-agent harness is direct independent evidence that worker-model orchestration is being implemented and tested.
2026-07-20T06:17:15Z
grounded: known/medium — Scott already holds this architecture in “Scout–Senior Split” and “Model Barbell,” and implements related task-aware routing in active agent systems. A reproduc
2026-07-20T06:14:57Z
case created — Published benchmark figures and a hands-on Fable-plus-GPT workflow make the cost-performance claim directly reproducible.