fix(models): default fable-5 to adaptive thinking when no effort is set

[posthog]

Problem

Users who pick the new claude-fable-5 model hit hard failures whenever a generation runs without a configured effort/thinking level — 8 users and 142 failed generations on 06-11 alone, all invisible to error monitoring ($ai_http_status=None, so they never trip >= 400 alerts).

Since 06-10 these generations are deterministically rejected by Anthropic with invalid_request_error: "thinking.type.disabled" is not supported for this model. The effort level only reaches the SDK when present, so a fable-5 session with no effort set lets the SDK/CLI default to thinking: { type: "disabled" }, which fable-5 (adaptive-only) rejects outright.

Why: fable-5 was made user-selectable again by the 06-10 revert and only supports adaptive/enabled thinking — but our effort→options mapping leaves thinking disabled when no effort is configured, breaking ~2.6% of fable-5 generations.

Changes

• Add a requiresAdaptiveThinking capability flag in session/models.ts (alongside the existing MODELS_WITH_EFFORT set), with claude-fable-5 as the only current member.
• In the effort→options mapping in session/options.ts, default these models to a shared DEFAULT_EFFORT ("medium") when no effort is configured, so the SDK requests thinking: { type: "adaptive" } + output_config: { effort } instead of disabled thinking. All other models keep their existing behavior (an explicit effort is still forwarded unchanged).
• Thread the resolved gateway modelId into buildSessionOptions to gate the above.

How did you test this?

• Added unit tests in models.test.ts (requiresAdaptiveThinking matches only fable-5) and options.test.ts (fable-5 defaults to medium; explicit effort forwarded unchanged; opus-4-8 and no-model-id leave effort unset). Ran pnpm --filter @posthog/agent on both files — all pass.
• pnpm --filter @posthog/agent typecheck passes; Biome clean; host-boundary check shows no new violations.

Automatic notifications

• Publish to changelog?
• Alert Sales and Marketing teams?

---

Created with PostHog Code from an inbox report.

[github-actions]

React Doctor found no issues in the changed files. 🎉

_{Reviewed by React Doctor for commit 2af9068.}

[greptile-apps]

Prompt To Fix All With AI

Fix the following 1 code review issue. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 1
packages/agent/src/adapters/claude/session/models.test.ts:67-73
The team prefers parameterised tests, and this test checks the same `requiresAdaptiveThinking` function against five different model IDs with expected booleans — a perfect fit for `it.each`. Without parameterisation, a future maintainer adding a new model would need to know to add a new `expect` line inside an existing test rather than extend a table, and a single failure obscures which model caused it.

```suggestion
  it.each([
    ["claude-fable-5", true],
    ["claude-opus-4-8", false],
    ["claude-opus-4-7", false],
    ["claude-sonnet-4-6", false],
    ["claude-haiku-4-5", false],
  ] as const)(
    "requiresAdaptiveThinking(%s) === %s",
    (modelId, expected) => {
      expect(requiresAdaptiveThinking(modelId)).toBe(expected);
    },
  );
```

_{Reviews (1): Last reviewed commit: "fix(models): default fable-5 to adaptive..." | Re-trigger Greptile}

[greptile-apps]

The team prefers parameterised tests, and this test checks the same requiresAdaptiveThinking function against five different model IDs with expected booleans — a perfect fit for it.each. Without parameterisation, a future maintainer adding a new model would need to know to add a new expect line inside an existing test rather than extend a table, and a single failure obscures which model caused it.

Suggested change

	it("flags only models that reject disabled thinking as requiring adaptive thinking", () => {
	expect(requiresAdaptiveThinking("claude-fable-5")).toBe(true);
	expect(requiresAdaptiveThinking("claude-opus-4-8")).toBe(false);
	expect(requiresAdaptiveThinking("claude-opus-4-7")).toBe(false);
	expect(requiresAdaptiveThinking("claude-sonnet-4-6")).toBe(false);
	expect(requiresAdaptiveThinking("claude-haiku-4-5")).toBe(false);
	});
	it.each([
	["claude-fable-5", true],
	["claude-opus-4-8", false],
	["claude-opus-4-7", false],
	["claude-sonnet-4-6", false],
	["claude-haiku-4-5", false],
	] as const)(
	"requiresAdaptiveThinking(%s) === %s",
	(modelId, expected) => {
	expect(requiresAdaptiveThinking(modelId)).toBe(expected);
	},
	);

Context Used: Do not attempt to comment on incorrect alphabetica... (source)

Prompt To Fix With AI

This is a comment left during a code review.
Path: packages/agent/src/adapters/claude/session/models.test.ts
Line: 67-73

Comment:
The team prefers parameterised tests, and this test checks the same `requiresAdaptiveThinking` function against five different model IDs with expected booleans — a perfect fit for `it.each`. Without parameterisation, a future maintainer adding a new model would need to know to add a new `expect` line inside an existing test rather than extend a table, and a single failure obscures which model caused it.

```suggestion
  it.each([
    ["claude-fable-5", true],
    ["claude-opus-4-8", false],
    ["claude-opus-4-7", false],
    ["claude-sonnet-4-6", false],
    ["claude-haiku-4-5", false],
  ] as const)(
    "requiresAdaptiveThinking(%s) === %s",
    (modelId, expected) => {
      expect(requiresAdaptiveThinking(modelId)).toBe(expected);
    },
  );
```

**Context Used:** Do not attempt to comment on incorrect alphabetica... ([source](https://app.greptile.com/review/custom-context?memory=instruction-0))

How can I resolve this? If you propose a fix, please make it concise.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!