Skip to content

The ultimate harness for coding agents and beyond. All your agents. All your machines. One command center. Start with code, then follow your curiosity and build across disciplines: CAD, circuits, robots, games and music.

Routing template: small work on Sonnet or Sol at high effort, big work on Opus or Astra (#2495)

0xken150ee0c46ffa7e
diff --git a/backend/src/superTabs/catalog.test.ts b/backend/src/superTabs/catalog.test.ts
index 80d76e104..06ceb5cb0 100644
--- a/backend/src/superTabs/catalog.test.ts
+++ b/backend/src/superTabs/catalog.test.ts
@@ -4,7 +4,7 @@ import { DEFAULT_ROUTING_POLICY } from './routingPolicy.js'
import { BENCHMARK_PRIORS } from './benchmarkPriors/data.js'
import { PRICES } from './costs/prices.js'
-const NOT_OFFERED = 'Not offered by default: no Engineer runs on Claude Sonnet or Claude Haiku (Ken and Dee, 2026-10-30).'
+const NOT_OFFERED = 'Not offered by default: no agent runs on Claude Haiku, and none below high effort (Ken, 2026-11-02).'
describe('the model catalog', () => {
it('lists today\'s models with their strength', () => {
@@ -14,10 +14,11 @@ describe('the model catalog', () => {
{ engine: 'codex', model: 'default', effort: 'default', strength: 3, cost: 2, offered: true },
{ engine: 'codex', model: 'gpt-6.1-sol', effort: 'high', strength: 3, cost: 2, offered: true },
{ engine: 'codex', model: 'gpt-6-astra', effort: 'xhigh', strength: 3, cost: 5, offered: true },
- { engine: 'claude', model: 'sonnet', effort: 'medium', strength: 2, cost: 2, offered: false, note: NOT_OFFERED },
+ { engine: 'claude', model: 'sonnet', effort: 'high', strength: 2, cost: 2, offered: true },
{ engine: 'claude', model: 'haiku', effort: 'auto', strength: 1, cost: 1, offered: false, note: NOT_OFFERED },
])
- expect(MODEL_CATALOG.models.filter(m => m.offered).every(m => m.strength === 3)).toBe(true)
+ // Ken, 2026-11-02: small work runs on Sonnet or Sol, nothing below high effort; Haiku stays off.
+ expect(MODEL_CATALOG.models.filter(m => m.offered).every(m => m.strength >= 2 && ['high', 'xhigh', 'default'].includes(m.effort))).toBe(true)
expect(MODEL_CATALOG.grid).toEqual({ strength: 1, models: [{ model: 'kimi k3', strength: 2 }, { model: 'deepseek v4.1', strength: 2 }] })
})
diff --git a/backend/src/superTabs/catalog.ts b/backend/src/superTabs/catalog.ts
index fd32acecf..eba09f62e 100644
--- a/backend/src/superTabs/catalog.ts
+++ b/backend/src/superTabs/catalog.ts
@@ -18,7 +18,7 @@ export interface CatalogModel { engine: 'claude' | 'codex'; model: string; effor
* `strength` (the CLI's route, superTabs/route.ts gridStrength). */
export interface CatalogGrid { strength: number; models: ReadonlyArray<{ model: string; strength: number }> }
-const NOT_OFFERED = 'Not offered by default: no Engineer runs on Claude Sonnet or Claude Haiku (Ken and Dee, 2026-10-30).'
+const NOT_OFFERED = 'Not offered by default: no agent runs on Claude Haiku, and none below high effort (Ken, 2026-11-02).'
export const MODEL_CATALOG: { models: readonly CatalogModel[]; grid: CatalogGrid } = {
models: [
@@ -30,10 +30,12 @@ export const MODEL_CATALOG: { models: readonly CatalogModel[]; grid: CatalogGrid
// Codex's own default model, named: GPT-6.1 Sol scores 58.18 on Terminal-Bench 4.0 in Codex (benchmarkPriors/data.ts) and
// costs Sonnet's API rate (costs/prices.ts).
{ engine: 'codex', model: 'gpt-6.1-sol', effort: 'high', strength: 3, cost: 2, offered: true },
- // Builders run on Opus high or Astra xhigh (the owners' standing decision). Its API rate is five times Sol's and 2.5
+ // Big builds run on Opus high or Astra xhigh (the owners' standing decision). Its API rate is five times Sol's and 2.5
// times Opus's, so it costs 5, after Opus at xhigh: at equal strength the route keeps choosing Opus high first.
{ engine: 'codex', model: 'gpt-6-astra', effort: 'xhigh', strength: 3, cost: 5, offered: true },
- { engine: 'claude', model: 'sonnet', effort: 'medium', strength: 2, cost: 2, offered: false, note: NOT_OFFERED },
+ // Small work (a small feature, a fix, docs, tests) runs on Sonnet or Sol, at high effort, nothing below it (Ken,
+ // 2026-11-02; 2026-10-30 had turned Sonnet off). Big or risky work stays on Opus or Astra.
+ { engine: 'claude', model: 'sonnet', effort: 'high', strength: 2, cost: 2, offered: true },
{ engine: 'claude', model: 'haiku', effort: 'auto', strength: 1, cost: 1, offered: false, note: NOT_OFFERED },
],
// Grid models are free but weaker. Kimi K3 and DeepSeek v4.1 do Sonnet's work (Ken, 2026-10-29); any other counts 1.
diff --git a/backend/src/superTabs/instructions.test.ts b/backend/src/superTabs/instructions.test.ts
index b299ca691..338b8603a 100644
--- a/backend/src/superTabs/instructions.test.ts
+++ b/backend/src/superTabs/instructions.test.ts
@@ -36,7 +36,10 @@ describe('the built-in template', () => {
}
expect(INSTRUCTIONS_TEMPLATE).toContain('Review and test run in parallel')
expect(INSTRUCTIONS_TEMPLATE).toContain('Does not read the code')
- expect(INSTRUCTIONS_TEMPLATE).toContain('No Engineer runs on Claude Sonnet or Claude Haiku')
+ // Ken, 2026-11-02: small work on Sonnet or Sol, big work on Opus or Astra, nothing below high effort.
+ expect(INSTRUCTIONS_TEMPLATE).toContain('Claude Sonnet at high effort or Codex GPT-6.1 Sol at high effort')
+ expect(INSTRUCTIONS_TEMPLATE).toContain('No agent runs below high effort, and none on Claude Haiku.')
+ expect(INSTRUCTIONS_TEMPLATE).not.toContain('No Engineer runs on Claude Sonnet')
expect(INSTRUCTIONS_TEMPLATE).toMatch(/Low-risk docs, tests and fixtures go to a model on the Swarm's grid first/)
expect(INSTRUCTIONS_TEMPLATE).toContain('Does not write the code itself')
// A Yes to a nightly routing proposal is a line the PM writes into Routing (project instructions, phase 4, decision 21).
diff --git a/backend/src/superTabs/instructionsTemplate.ts b/backend/src/superTabs/instructionsTemplate.ts
index 91ff3da01..a09acc621 100644
--- a/backend/src/superTabs/instructionsTemplate.ts
+++ b/backend/src/superTabs/instructionsTemplate.ts
@@ -57,11 +57,14 @@ export const TEMPLATE_WORKFLOW = `## Workflow
export const TEMPLATE_ROUTING = `## Routing
-- Engineers run on Claude Opus at high effort or Codex GPT-6 Astra at xhigh, whichever has quota at the better price. No
- engine is preferred. No Engineer runs on Claude Sonnet or Claude Haiku (Ken and Dee, 2026-10-30).
+- Small work (a small feature, a fix, docs, tests) runs on
+ Claude Sonnet at high effort or Codex GPT-6.1 Sol at high effort, whichever has quota at the better price.
+- Big, hard or risky work runs on Claude Opus at high effort or Codex GPT-6 Astra at xhigh, whichever has quota at the
+ better price. No engine is preferred.
+- No agent runs below high effort, and none on Claude Haiku.
- Reviewers run on a model at least as strong as the Engineer's, on either engine, never on a model of the Swarm's grid.
-- Low-risk docs, tests and fixtures go to a model on the Swarm's grid first. After a failed review they move to Opus or
- Astra.
+- Low-risk docs, tests and fixtures go to a model on the Swarm's grid first. After a failed review they move to Sonnet or
+ Sol, and after a second to Opus or Astra.
- A Tester runs on a machine that has what the change needs: a Mac for Desktop, the machine the device is plugged into for
firmware. The machine comes before the model. A mid-strength model is enough for most testing.`
diff --git a/backend/src/superTabs/migration/instructions.test.ts b/backend/src/superTabs/migration/instructions.test.ts
index 2ad418c81..c4a388268 100644
--- a/backend/src/superTabs/migration/instructions.test.ts
+++ b/backend/src/superTabs/migration/instructions.test.ts
@@ -32,10 +32,22 @@ describe('the document the migration writes', () => {
expect(text).toContain('- Builder runs on Codex gpt-6-astra at xhigh effort unless its start names another.')
})
+ // Ken, 2026-11-02 turned Sonnet on at high effort in the default table. A policy saved before kept the old default
+ // row (Sonnet medium, off): that is the old default, not the owner's choice, and says nothing.
+ it('reads a policy saved with the old default Sonnet row as no difference', () => {
+ const policy: RoutingPolicy = { ...DEFAULT_ROUTING_POLICY,
+ models: [...DEFAULT_ROUTING_POLICY.models.filter(m => m.model !== 'sonnet'),
+ { engine: 'claude', model: 'sonnet', effort: 'medium', quality: 2, cost: 2, enabled: false }] }
+ const { text } = documentFor({ project, roleTexts: {}, rules: [], policy })
+ expect(text.slice(INSTRUCTIONS_TEMPLATE.trimEnd().length)).not.toContain('sonnet')
+ })
+
it('adds the project\'s rules in the owner\'s words, and the policy\'s notes and differences as plain lines, under Routing', () => {
const policy: RoutingPolicy = { ...DEFAULT_ROUTING_POLICY,
- models: [...DEFAULT_ROUTING_POLICY.models.filter(m => !(m.engine === 'claude' && m.model === 'haiku'))
- .map(m => m.model === 'sonnet' ? { ...m, enabled: true } : m.model === 'opus' && m.effort === 'high' ? { ...m, cost: 5 } : m),
+ // Saved before 2026-11-02: Sonnet's row was medium effort, which this owner had turned on; no Sonnet high row.
+ models: [...DEFAULT_ROUTING_POLICY.models.filter(m => !(m.engine === 'claude' && (m.model === 'haiku' || m.model === 'sonnet')))
+ .map(m => m.model === 'opus' && m.effort === 'high' ? { ...m, cost: 5 } : m),
+ { engine: 'claude', model: 'sonnet', effort: 'medium', quality: 2, cost: 2, enabled: true },
{ engine: 'codex', model: 'gpt-7', effort: 'high', quality: 2, cost: 9, enabled: true }],
floor: { low: 1, medium: 3, high: 3 }, quotaReserve: 20, notes: 'Keep Opus for hard work.',
localModels: { quality: 2, enabled: false, models: [{ model: 'kimi k3', quality: 3 }] },
@@ -48,6 +60,8 @@ describe('the document the migration writes', () => {
expect(added).not.toContain('Another project')
expect(added).toContain('- Routing notes: Keep Opus for hard work.')
expect(added).toContain('- Claude sonnet at medium effort is on.')
+ // The Sonnet high row came after the policy was saved: its absence is no choice of the owner's.
+ expect(added).not.toContain('sonnet at high effort')
expect(added).toContain('- Claude haiku at auto effort is not offered.')
expect(added).toContain('- Claude opus at high effort counts as strength 3 of 3 at cost 5.')
expect(added).toContain('- Codex gpt-7 at high effort is offered at strength 2 of 3, cost 9.')
diff --git a/backend/src/superTabs/migration/instructions.ts b/backend/src/superTabs/migration/instructions.ts
index 29e935993..d06d5cd02 100644
--- a/backend/src/superTabs/migration/instructions.ts
+++ b/backend/src/superTabs/migration/instructions.ts
@@ -43,9 +43,19 @@ const rowName = (row: Pick<RoutingModel, 'engine' | 'model' | 'effort'>) =>
const sameRow = (a: Pick<RoutingModel, 'engine' | 'model' | 'effort'>, b: Pick<RoutingModel, 'engine' | 'model' | 'effort'>) =>
a.engine === b.engine && a.model === b.model && a.effort === b.effort
+/** Default rows a saved policy may still hold from before the default table changed: the old row as it was, and the row
+ * that replaced it. Ken, 2026-11-02 turned Sonnet on at high effort; until then the default row was Sonnet at medium,
+ * off. A policy saved before holds the old row unchanged and lacks the new one, and neither is a choice of its owner's. */
+const REPLACED_DEFAULTS: ReadonlyArray<{ old: RoutingModel; now: Pick<RoutingModel, 'engine' | 'model' | 'effort'> }> = [
+ { old: { engine: 'claude', model: 'sonnet', effort: 'medium', quality: 2, cost: 2, enabled: false }, now: { engine: 'claude', model: 'sonnet', effort: 'high' } },
+]
+
/** The policy's settings that differ from the built-in policy, each as one sentence. */
function policyLines(policy: RoutingPolicy): string[] {
- const base = effectivePolicy(null, null).policy, lines: string[] = []
+ const current = effectivePolicy(null, null).policy, lines: string[] = []
+ // A policy saved with a replaced default row is read against the default it was saved with.
+ const replaced = REPLACED_DEFAULTS.filter(r => policy.models.some(m => sameRow(m, r.old)) && !policy.models.some(m => sameRow(m, r.now)))
+ const base = { ...current, models: [...current.models.filter(m => !replaced.some(r => sameRow(m, r.now))), ...replaced.map(r => r.old)] }
if (policy.notes.trim()) lines.push(`Routing notes: ${policy.notes.trim()}`)
for (const row of policy.models) {
const built = base.models.find(m => sameRow(m, row))
diff --git a/backend/src/superTabs/routingPolicy.test.ts b/backend/src/superTabs/routingPolicy.test.ts
index e473e2281..36a481763 100644
--- a/backend/src/superTabs/routingPolicy.test.ts
+++ b/backend/src/superTabs/routingPolicy.test.ts
@@ -15,23 +15,24 @@ describe('routing policy', () => {
expect(DEFAULT_ROUTING_POLICY.models).toContainEqual({ engine: 'claude', model: 'opus', effort: 'xhigh', quality: 3, cost: 4, enabled: true })
})
- // Ken, 2026-10-30: no Sonnet Builders. Only the default table changes; an owner's own table is theirs (Lead's ruling).
- it('turns Sonnet off in the default table, and keeps an owner\'s saved Sonnet row and a tab override\'s as they are', () => {
+ // Ken, 2026-11-02: Sonnet at high effort is back for small work (2026-10-30 had turned it off). Only the default table
+ // changes; an owner's own table is theirs (Lead's ruling).
+ it('offers Sonnet at high effort in the default table, and keeps an owner\'s saved Sonnet row and a tab override\'s as they are', () => {
const sonnet = (models: Array<{ engine: string; model: string; enabled: boolean }>) => models.filter(m => m.engine === 'claude' && m.model === 'sonnet')
- expect(sonnet(DEFAULT_ROUTING_POLICY.models)).toEqual([expect.objectContaining({ effort: 'medium', quality: 2, enabled: false })])
+ expect(sonnet(DEFAULT_ROUTING_POLICY.models)).toEqual([expect.objectContaining({ effort: 'high', quality: 2, enabled: true })])
// Haiku too (review of #2314): an older CLI on the new server picked it for easy work the server refuses, on every route.
expect(DEFAULT_ROUTING_POLICY.models.filter(m => m.model === 'haiku')).toEqual([expect.objectContaining({ quality: 1, enabled: false })])
- // What an older CLI's route can still pick from the default table builds: every row it may pick is rated 3.
- expect(DEFAULT_ROUTING_POLICY.models.filter(m => m.enabled).every(m => m.quality === 3)).toBe(true)
- expect(sonnet(effectivePolicy(null, null).policy.models).map(m => m.enabled)).toEqual([false])
- // An owner who saved the table before keeps Sonnet on; it is their row.
- const saved = { ...DEFAULT_ROUTING_POLICY, models: DEFAULT_ROUTING_POLICY.models.map(m => m.model === 'sonnet' ? { ...m, enabled: true } : m) }
+ // Nothing it may pick is rated below 2 (an older CLI's own route still keeps its builds at 3).
+ expect(DEFAULT_ROUTING_POLICY.models.filter(m => m.enabled).every(m => m.quality >= 2)).toBe(true)
+ expect(sonnet(effectivePolicy(null, null).policy.models).map(m => m.enabled)).toEqual([true])
+ // An owner who saved the table with Sonnet off keeps it off; it is their row.
+ const saved = { ...DEFAULT_ROUTING_POLICY, models: DEFAULT_ROUTING_POLICY.models.map(m => m.model === 'sonnet' ? { ...m, enabled: false } : m) }
const read = effectivePolicy(saved, null)
expect(read.warnings).toEqual([])
- expect(sonnet(read.policy.models).map(m => m.enabled)).toEqual([true])
+ expect(sonnet(read.policy.models).map(m => m.enabled)).toEqual([false])
expect(read.policy.models.filter(m => m.model !== 'sonnet')).toEqual(saved.models.filter(m => m.model !== 'sonnet'))
// A tab override's own models win over the default too.
- expect(sonnet(effectivePolicy(null, { models: saved.models }).policy.models).map(m => m.enabled)).toEqual([true])
+ expect(sonnet(effectivePolicy(null, { models: saved.models }).policy.models).map(m => m.enabled)).toEqual([false])
})
it('uses the default when the account has none, and merges a tab override field by field', () => {
expect(effectivePolicy(null, null)).toEqual({ policy: DEFAULT_ROUTING_POLICY, warnings: [], overridden: [] })