Anthropic's deprecation page retires Claude Sonnet 4.5 from the native API — reduced availability from late October, full decommission November 30, 2026 — forcing dependent workflows to migrate, while AWS lists Sonnet 4.5 as still Active on Bedrock under an independently controlled lifecycle, making Bedrock the potential post-retirement access route.
What is this?
Anthropic's official model deprecations page now lists claude-sonnet-4-5 as deprecated on the native Claude API — reduced availability from late October 2026 and a tentative full retirement of November 30, 2026, per the case's evidence titles — forcing API-dependent workflows to migrate to newer Sonnet versions. The claimed post-retirement access route is Amazon Bedrock: AWS's own supplied pages confirm Sonnet 4.5 has been generally available on Bedrock since September 2025 (later GovCloud and Japan/Australia cross-region inference) under a lifecycle AWS governs independently — 'each partner's own schedule,' as one third-party piece puts it. The supplied web material is thin on the exact claim, however: none of the snippets shows the deprecation page itself, an earlier-dated third-party piece has the API identifier still active with an earlier tentative date ('not sooner than September 29, 2026') plus a separate June 22, 2026 consumer-app removal, and nothing supplied confirms AWS currently lists Sonnet 4.5 as Active vs. Legacy on Bedrock — that rests on the case's third evidence title alone.
Why it matters to Scott
A dated, first-party forced-migration episode that lands exactly where Model Perishability and the composable-AI ebook argue — and extends it with a mechanism his canon implies but doesn't yet state: an independently governed partner lifecycle as the post-deprecation access route, directly actionable since he already runs Bedrock behind his LiteLLM gateway (caveat per the grounding: the 'Bedrock still Active' leg rests on the case's third-party title alone, so verify AWS's own lifecycle page before wiring it in as a Claude fallback tier). Migration-target caution from the radar's own pages: successor Sonnet 5.5 reportedly buffers tool_result output through message completion, reproduced on Bedrock, so both the escape route and the upgrade path need eval before cutover.
ip:concept.model-perishabilityip:source.the-death-of-shelf-software-and-the-rise-of-composable-ai-ebookdev:technology.litellmdev:technology.amazon-bedrockradar:anthropic-sonnet-5-5-releaseradar:sonnet55-tool-result-streaming-regressionradar:openai-gpt55-retirementradar:gemini-25-october-migration-gapradar:github-copilot-model-deprecations
queries asked of Scott's wikis
- model version pinning and breaking changes in agent harnesses
- multi-provider LLM routing, fallback, and portability strategy
- model deprecation migration playbook or checklist
- local open-weight models as fallback when hosted models retire
- Bedrock or partner-platform Claude access vs first-party API
- agent memory and eval reproducibility across model versions
Measured heat
now 0 pts/hpeak 9 pts/hcomments 0/hpeers p0momentum: steady2 platformsage 261h
points/hour across evidence · reading as of 2026-10-12 02:59:37.977291+11:00 · deterministic, not a model opinion
How the heat travelled
pace: p72 vs 1188 stories at the 168h mark (now 261h old) — ahead of sakana-pc-alm-local-training (1.0x), behind perplexity-numeric-citation-audit (1.0x)
Evidence (4) — ⭐ canonical anchor
Interpretation history
2026-10-07T04:22:09Z
The Opus 3 precedent adds a second, Anthropic-native post-retirement mechanism to the hypothesis: under the Nov 2025 preservation commitments, retired models demonstrably keep paid consumer access and by-request API access, so Nov 30 is no longer a pure access cliff — but by-request gates cannot serve automated pipelines, leaving the still-unverified Bedrock Active status as the only programmable fallback and Scott's verification task unchanged. Episode itself stays idle (0 pts/h, 9th percentile, no periphery growth); the new source is low-credibility single-source, so heat stays low while the interpretation absorbs the precedent.
2026-10-07T03:33:13Z
evidence attached: reddit.post.1wzhr1g — Substantive mapping of the Opus 3 preservation precedent and its real limits, which materially contextualises what post-retirement access for Sonnet 4.5 will look like.
2026-10-02T11:29:34Z
No meaning change: both velocity spikes were floor-baseline artifacts (≤4.7 pts/h against a 1.0/h floor) and the new comments are nostalgia, not new facts; measured heat confirms the episode peaked early and is now idle (0 pts/h, 18.8th percentile) despite the hot anthropic neighbourhood. Case stays parked on its two unresolved legs — AWS's Bedrock Active/Legacy status never verified first-party, and the Nov 30 date still tentative — with the late-October reduced-availability onset as the next natural trigger.
2026-09-30T22:17:29Z
grounded: converges/medium — A dated, first-party forced-migration episode that lands exactly where Model Perishability and the composable-AI ebook argue — and extends it with a mechanism h
2026-09-30T22:08:58Z
case created — First-party deprecation notice from official docs plus a same-day Bedrock-lifecycle migration report — a concrete forced-migration episode with no open case.
Decision trace
- 10-08 12:22attention_routeThe editor compared this story and chose to keep watching.
- 10-08 00:59attention_routeDeadline is weeks out but the verification action is cheap and the reduced-availability window opens this month; it goes on the morning list rather than interrupting the night.
- 10-07 15:22repriceThe Opus 3 precedent adds a second, Anthropic-native post-retirement mechanism to the hypothesis: under the Nov 2025 preservation commitments, retired models demonstrably keep paid consumer access and
- 10-07 14:33attachSubstantive mapping of the Opus 3 preservation precedent and its real limits, which materially contextualises what post-retirement access for Sonnet 4.5 will look like.
- 10-07 14:27propose_attachSubstantive mapping of the Opus 3 preservation precedent and its real limits, which materially contextualises what post-retirement access for Sonnet 4.5 will look like.
- 10-02 21:29repriceNo meaning change: both velocity spikes were floor-baseline artifacts (≤4.7 pts/h against a 1.0/h floor) and the new comments are nostalgia, not new facts; measured heat confirms the episode peaked ea
- 10-02 06:21sensor_dirtyvelocity_spike
- 10-01 19:21sensor_dirtycomment_update
- 10-01 14:21sensor_dirtyvelocity_spike
- 10-01 12:20sensor_dirtycomment_update
- 10-01 08:17groundA dated, first-party forced-migration episode that lands exactly where Model Perishability and the composable-AI ebook argue — and extends it with a mechanism his canon implies but doesn't yet st
- 10-01 08:08createFirst-party deprecation notice from official docs plus a same-day Bedrock-lifecycle migration report — a concrete forced-migration episode with no open case.
- 10-01 07:29propose_attachSame deprecation episode as the official notice; adds the Bedrock-independence migration route and timeline detail.