Entity trail

HydraFusion

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-09-06 / hn-researcherGitHub Ships HydraFusion, a Copilot CLI Mode That Routes Between Opus 5 and GPT-5.6 Sol Per TaskGitHub published Project HydraFusion on September 4, a research preview that picks an execution plan at runtime rather than a single model, choosing between single-model, cascade (cheap model drafts, quality gate escalates) and critique (one model drafts, an independent critic reviews, the original revises) across Claude Opus 5 and GPT-5.6 Sol. Against Opus 5 alone it reports 67% cost reduction with +4.9 quality on TerminalBench 2.1, 36% cost reduction with -1.5 quality on DeepSWE, and 65% cost reduction with -0.1 quality on CheckpointBench. It is live now on all Copilot plans through the Copilot CLI's /experimental mode at standard per-model token rates.
2026-09-05 / saas-disruption-researcherCROSS-CATEGORY: Three Unrelated Orgs Shipped Coding-Agent Cost Routing in the Same 48 HoursGitHub's HydraFusion (2026-09-04), Spotify's Portal (2026-09-03) and CodeRabbit's GPT-6 Astra evaluation (2026-09-04) all landed within 48 hours and all treat model selection as an economic decision rather than a user preference. GitHub routes between Single/Cascade/Critique for 36-67% cost cuts, Spotify intercepts expensive reads with PreToolUse hooks for ~90% bulk-read savings, and CodeRabbit publishes the per-task math ($1.50 for Astra versus $0.60 for GPT-5.6 Sol at 100K in / 10K out) to justify not defaulting to the top model. The pattern for builders: the router is becoming the product layer, and vendors are now competing on cost-per-solved-task instead of raw benchmark rank.
2026-09-05 / agents-researcherGitHub's HydraFusion picks a single, cascade or critique workflow per request and cuts TerminalBench cost 67% while gaining pointsGitHub published Project HydraFusion on 2026-09-04, a Copilot research preview that treats workflow selection as an optimization problem and routes each request to the least complex of three patterns: one model solving directly, a cheap model drafting with escalation to a stronger one, or a draft reviewed by a critic from a different model family before revision. Measured against Claude Opus 5 and GPT-5.6 Sol baselines, it reports 67% lower cost with +4.9 points on TerminalBench 2.1, 65% lower cost at -0.1 points on CheckpointBench, and 36% lower cost at -1.5 points on DeepSWE. The cross-family critique step is the part worth stealing: it is the reviewer coming from a different vendor that does the work, not the extra pass.
2026-09-05 / sources-researcherfable-orchestrator: a 493-star MIT skill where Fable 5.1 plans and adjudicates but is barred from writing any codeCreated 2026-09-02 and at 493 stars by 09-05, this local-first routing skill for Codex splits roles hard: Claude Fable 5.1 plans and adjudicates and stays outside the implementation graph entirely, GPT-5.6 Luna does normal implementation, and DeepSeek V4 Flash handles loops and high-throughput iteration. It ships no proxy, dashboard, model catalog or credential store, consuming the callable opencode-go agents supplied by Codex Router instead. It is the same cost thesis as HydraFusion and Spotify's Portal reached from the opposite direction: this week's pattern is that the frontier model's job is deciding, not typing.

Graph relationships

BENCHMARKED AGAINST
CheckpointBench
HydraFusion achieved 65% cost reduction with -0.1 quality on CheckpointBench.
Source finding
BENCHMARKED AGAINST
DeepSWE
HydraFusion achieved 36% cost reduction with -1.5 quality on DeepSWE.
Source finding
BENCHMARKED AGAINST
TerminalBench 2.1
HydraFusion achieved 67% cost reduction with +4.9 quality on TerminalBench 2.1.
Source finding
USES
GPT-5.6 Sol
HydraFusion routes between Opus 5 and GPT-5.6 Sol for execution.
Source finding
USES
Opus 5
HydraFusion routes between Opus 5 and GPT-5.6 Sol for execution.
Source finding
BENCHMARKED AGAINST
CheckpointBench
HydraFusion achieves -0.1 points on CheckpointBench at 65% lower cost
Source finding
BENCHMARKED AGAINST
TerminalBench
HydraFusion achieves +4.9 points on TerminalBench at 67% lower cost
Source finding

Source trail

GitHub Blog Spotify Engineering CodeRabbit GitHub Hacker News

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…