Anthropic has not announced Claude Opus 5.2. As of September 18, 2026, we found no Opus 5.2 entry and no claude-opus-5-2 identifier in Anthropic's model catalog, release notes, pricing page, or SDK model lists, and EvoLink has no route. The name comes from community reports of an unlabeled test inside Claude Code. Join the alert to get the verified EvoLink model ID, live price, and first-call guidance when a real route exists; until then, Claude Opus 5 is the documented Opus model you can use today.
We will email you only after Anthropic publishes the model and EvoLink verifies a callable route, returned model identity, price, usage accounting, and billing, not when another routing screenshot appears.
EvoLink product alerts only. This is not an Anthropic program and not an access guarantee.
Want faster updates? Join the EvoLink Discord
The update will include the exact EvoLink model ID, live price, supported effort and thinking controls, access limits, first-call guidance, and a migration path that keeps Claude Opus 5 as the measured fallback.
The status above comes from first-party Anthropic and cloud-provider records checked on September 17–18, 2026, not from routing screenshots or release-date predictions.
The models overview lists Claude Fable 5.1, Claude Opus 5, Claude Sonnet 5, and Claude Haiku 4.5. The newest Opus release note is Claude Opus 5 on July 24, 2026. No Opus 5.2 product, preview, or system card exists as of September 18, 2026.
Community posts since September 14 describe an unlabeled routing test inside Claude Code; one widely shared post says web requests still reach the older model. Anthropic has not confirmed a test, its scope, or the name of the build. Reports also disagree on which plans are included.
Anthropic's current ID format is why people expect "claude-opus-5-2". We did not find that string in the API model docs, the Python or TypeScript SDK model lists, the Claude Code model-config docs, or the Claude Code 2.1.276 binary, which does contain claude-opus-5 and claude-fable-5-1. A naming format does not commit Anthropic to a product or an ID.
We found no Opus 5.2 model in the three cloud catalogs. The "claude-opus-5-2" slug circulating on X traces to a community-maintained registry file whose values equal Claude Opus 5; the release tracker links the file and explains what that does and does not establish.
EvoLink has no Opus 5.2 route, SKU, or integration date. This page switches from availability watch to live access only after route identity, request success, usage, price, and billing are verified. Claude Opus 5 is listed today.
Lists claude-opus-5 as the current Opus model with $5 input / $25 output per MTok, 1M context, and 128K max output. No Opus 5.2 row.
The newest Opus entry is the July 24, 2026 Claude Opus 5 launch; the September 1 entry covers Claude Fable 5.1 and Mythos 5.1 only.
Documents that Claude Opus 5 already supports all five effort levels (low, medium, high, xhigh, max). A five-level effort list is therefore not evidence of a new model.
Lists the model IDs Claude Code accepts, ending at claude-opus-5 for the Opus line. It does not mention Opus 5.2 or any routing test.
The per-model table (updated September 11, 2026) lists claude-opus-5 and defines "version 2" as Azure-hosted deployments. It has no claude-opus-5-2 row.
States that model weights are fixed for a given ID, that an updated version ships under a new ID, and that serving-infrastructure changes are the most likely cause of behavior differences on a stable ID.
This page changes from availability watch to live access only after Anthropic publishes the model and its API contract and EvoLink independently verifies a callable route, returned identity, usage, and billing. A routing report, a registry file, or an Anthropic announcement alone does not change the EvoLink status.
September 14–17 update
Routing reports increased; the official status is unchanged
X users report faster, less "lazy" Opus 5 output inside Claude Code and attribute it to a hidden Opus 5.2 build; reports disagree on whether the web app, Pro plans, or Max plans are included, and some say the change was withdrawn again. None of the posts include a model identifier, response metadata, or an Anthropic statement, so they do not establish a public API, a model ID, a price, or EvoLink access.
Faster answers, shorter output, a trivia question answered correctly, or a new slug in a third-party file are not identity proof. Verification means the public product, the documented ID, the returned metadata, and the bill all agree.
Treat the September routing reports as investigation leads, not as availability. A real Opus 5.2 API claim needs a documented model ID, returned identity, channel scope, price, usage, and repeatable request behavior.
"Opus 5.2" is a community label, and we did not find a matching ID in any official source. The August "Opus 5.1" reports never became a product under that name; Claude Fable 5.1 was released instead.
The popular "ask who Tibo is, no search" check reads generated text, and results reportedly differ across Claude Code versions on the same machine. Returned metadata is supplied by the server, so it is not independent proof of the weights either; what you can check is that it agrees with the documented ID mapping and with billing.
Users report different behavior in Claude Code, the web app, and third-party agents, and on Pro versus Max plans. A consumer-plan observation does not establish an API entitlement, a price, or an EvoLink route.
Some users report the faster behavior disappeared again within a day. A staged test that can be withdrawn proves nothing about capacity, quotas, or whether the route is fit for production.
Prepare evidence, configuration, and rollback paths now so launch-day work is validation instead of a rewrite. Past Opus releases have changed defaults, token accounting, and accepted parameters.
Do not hard-code claude-opus-5-2. Leave the candidate unset until EvoLink verifies a callable route and returned model identity.
Collect 20–50 real coding, agent, tool, and knowledge-work traces with objective acceptance criteria, including the long tasks where Opus 5 asked you to continue.
Record quality, latency, tool errors, tokens, cache hits, reviewer time, and cost per accepted task at the effort levels you actually use (high and xhigh for most agent work).
Set thresholds per task class for pass rate, p95 latency, API cost per accepted task, human fix time, and rollback before the model exists, not during launch week.
Capture requested and returned model, effort, thinking state, tools, usage, stop reason, fallback chain, and billing. Opus 4.7 changed the tokenizer and Opus 5 changed thinking defaults; a point release can do the same.
Keep Claude Opus 5 serving users until the candidate proves a workload-specific improvement under controlled traffic, and test the Opus 5 route you would roll back to before you need it.
These are the transaction fields an API buyer needs. Unknown fields stay unknown until Anthropic documents them and EvoLink verifies its own route. Claude Opus 5 values are a predecessor reference, not an Opus 5.2 forecast. Current Opus 5.2 answer: product, model ID, price, limits, effort defaults, release date, and EvoLink route are all unpublished or unverified.
| Field | Current verified status | Required before activation |
|---|---|---|
| Official product | Not announced in any Anthropic source checked on September 17–18, 2026 | Dated Anthropic announcement, model page with a Released date, or release-notes entry |
| Public API model ID | Not published; claude-opus-5-2 is only this page's slug | Catalog entry, alias or snapshot behavior, and returned identity in response metadata |
| Provider pricing | Not published | Input, output, cache write and read, batch, fast-mode, and regional multipliers |
| Documented model to use today | Claude Opus 5 (claude-opus-5) at Anthropic's $5 input / $25 output per MTok base rates; 1M context; 128K max output; default effort high | Predecessor reference only; verify the EvoLink route and live billing before production |
| Effort and thinking controls | Opus 5 already supports low, medium, high, xhigh, and max; thinking cannot be disabled at xhigh or max | Whether defaults, allowed combinations, or token behavior change on the new model |
| EvoLink route and price | No active route or live price | Approved route, successful request, live price, usage accounting, and invoice match |
| Access scope | No documented channel, plan, account, region, or cloud scope; community reports disagree on plan coverage | Channel-specific availability and any tier, account, region, or capacity limits |
| Production verification | No Opus 5.2 route is available for matched testing | Returned identity, reliability, latency, cost per accepted task, fallback, and rollback |
Anthropic develops Claude; EvoLink's role is to make verified model access easier to compare, operate, switch, and roll back after a route is genuinely live. Today that means evaluating Claude Opus 5 and keeping model selection in configuration.
Keep one integration, key, billing surface, and observability path while the selected model changes through configuration.
Compare live route pricing and cost per accepted task only after request success, returned identity, usage, and billing are validated.
Evaluate the candidate offline and in canary traffic while retaining Claude Opus 5 as a measured rollback.
Use the release tracker for the dated evidence behind the routing reports and the comparison guide for what a point release would have to prove.
The September routing reports, the Foundry slug explained, the Opus 5.1 precedent, prediction-market odds, and the signals that would change this page.
For teams already on Opus 5: a task-by-task switch matrix, a worked cost-per-accepted-task example, a fillable acceptance record, and a tested way back.
The separate next-generation question. Opus 6 remains unannounced and is not the same rumor as the 5.2 routing reports.
No announcement was found in the sources checked on September 17–18, 2026: Anthropic's model catalog, release notes, pricing page, newsroom, and Claude Code changelog. The newest Opus release is Claude Opus 5 from July 24, 2026.
No. EvoLink has no Opus 5.2 route, SKU, or integration date. Join the alert, use the listed Claude Opus 5 route today, and verify it before production. The status changes only after Anthropic publishes the model and EvoLink verifies a successful request, returned identity, usage, and billing.
No model ID is published. People expect claude-opus-5-2 because of Anthropic's current ID format, but we did not find that string in any official documentation, SDK model list, or Claude Code build. It is this page's slug, not an identifier you can call; wait for the ID Anthropic publishes.
No provider or EvoLink price exists. Every release from Opus 4.5 through Opus 5 listed $5 input and $25 output per MTok, but that history is a reference point for your budget model, not an Opus 5.2 price.
Use Claude Opus 5 (claude-opus-5) for complex agentic coding and enterprise work, or Claude Fable 5.1 when Opus 5 at higher effort still falls short on demanding reasoning. Both are documented by Anthropic; verify the EvoLink route and live billing before production.
Anthropic lists claude-opus-5 as active with retirement not sooner than July 24, 2027 on the platforms it operates; Amazon Bedrock and Google Cloud set their own dates. That gives you an evaluation window, and it does not guarantee any particular gateway route, so test the Opus 5 route you plan to fall back to.
Verify the official name, the documented model ID, price, effort and thinking defaults, inputs, limits, tools, streaming, quotas, request success, returned metadata, usage, and billing, then move traffic by task class through replay, shadow, and canary stages with a rollback to Claude Opus 5 that you have actually tested.
No. It is a community label taken from X posts and a registry file. In August 2026, similar posts described an "Opus 5.1" that never shipped; Anthropic released Claude Fable 5.1 instead. Any future name is Anthropic's to announce.
Unverified. Several X users reported faster, less lazy Opus 5 behavior in Claude Code from September 14 and attributed it to a hidden build, and Anthropic has not commented. No post includes a model identifier or request record. Anthropic's versioning docs note that serving-infrastructure changes can alter behavior on a stable model ID.
No. Asking who "Tibo the reset guy" is reads generated text, not model identity, and users report different answers across Claude Code versions on the same machine. For API traffic, check that the documented ID mapping, the returned model field, and billing agree; those are consistency checks, not independent proof of the weights.
We did not find it in Microsoft's own Claude model table or the Foundry catalog's Anthropic page. The slug traces to a bot-generated file in a community model registry whose values equal Claude Opus 5; its -2 suffix lines up with Foundry's "version 2" label for Azure-hosted deployments. The release tracker links the file so you can check it.
Those levels already exist. Anthropic's effort documentation lists low, medium, high, xhigh, and max for Claude Opus 5, with xhigh introduced in Opus 4.7. A five-level effort list is not evidence of a new model.
Review provider-documented current models and verify each listed EvoLink route before production. Compare an Opus 5.2 candidate only after identity, request, usage, and billing verification.
Claude Opus 5
ListedAnthropic's current Opus model for repository-scale coding and long-running agents; the baseline any Opus 5.2 test must beat. Verify the EvoLink route before production.
View model
Claude Fable 5.1
ListedAnthropic's most capable listed model for demanding reasoning and long-horizon agent work; the model that shipped when "Opus 5.1" was rumored.
View model
Claude Model Family
CompareCompare current Claude routes by production role, model access, price, context, and workload fit.
View model