Claude Sonnet 4.8: What the March 2026 Leak Proved and Why the Model Never Shipped

On March 31, 2026, Anthropic published version 2.1.88 of the npm package @anthropic-ai/claude-code, according to VentureBeat with a 59.8 MB source map file that should never have become public. The reason: someone in the build process had forgotten to add *.map to the .npmignore. In a statement to VentureBeat, Anthropic called it “a release packaging issue caused by human error, not a security breach”. The result, according to analyses of the source code such as those by VentureBeat and InfoQ: roughly 512,000 lines of TypeScript source code in roughly 1,900 files, and thereby the largest accidental leak in the history of AI tools up to that point.

No Claude Sonnet 4.8 ever shipped. Anthropic took the Sonnet line from 4.6 straight to Claude Sonnet 5, released on 30 June 2026, which replaced Sonnet 4.6 as the default across its products the same day; Claude Opus 5 followed on 24 July 2026 and Opus 5.5 on 22 September 2026. Anthropic lists Sonnet 5 at 2 US dollars per million input and 10 dollars per million output tokens (API identifier claude-sonnet-5), below Sonnet 4.6 at 3/15 dollars, so the leaked-in price-parity assumption did not hold either. Claude Sonnet 5.5 followed on 28 September 2026 at the same price. The version string in the source map was real; the product inferred from it was not. That is exactly what this case shows: a source-map leak establishes that a name existed internally, not that it will ever be released.

After the leak, countless speculations about an allegedly imminent Claude Sonnet 4.8 circulated in tech forums, on X, and on English-language specialist blogs. The problem: most of these reports mixed verifiable facts from the leak with wishful thinking, and anyone who made budget or architecture decisions on that basis risked misinvestment. This article draws a strict line between what the source code objectively proved and what remained interpretation.

What the leak objectively reveals about Sonnet 4.8

The only verifiable hint of a model called "Sonnet 4.8" is found in the so-called Undercover Mode configuration of the Claude Code client. This mode activates automatically when an Anthropic employee (USER_TYPE === 'ant') uses Claude Code in a public repository that does not belong to Anthropic. Its function: it strips Co-Authored-By attributions from Git commits, blocks internal Slack channel names and — crucial for our analysis — forbids the mention of not-yet-released model versions.

In the list of forbidden strings that must be filtered in this Undercover Mode, the following entries appear:

• opus-4-7: released on 16 April 2026
• sonnet-4-8: never released; after Sonnet 4.6 came Sonnet 5
• mythos: released as Mythos Preview on 7 April 2026
• capybara: internal codename, according to VentureBeat and InfoQ for a Claude 4.6 variant
• numbat: unreleased at the time of the leak, unknown model

That is the entire concrete proof. The mere existence of the string sonnet-4-8 in an internal filter list proved that Anthropic used this version number internally, nothing more. In the entire leak there is no model card, no API documentation, no benchmarks, no feature flags that specifically reference Sonnet 4.8, and no migration function targeting this version string. The migration history in the code ends at migrateSonnet45ToSonnet46.

What the leak did not provide

The clarity of what we do not know is just as important as the facts themselves. In the context of AI roadmaps, half-information quickly tempts people into strategic missteps. Here is a list of what the leak did not provide:

No release date. Anthropic had communicated no timeline for Sonnet 4.8. The prediction market Manifold priced a release before May 24, 2026 at roughly 3 percent, so the market bet heavily against a near-term release. The pattern of past releases (Sonnet 4.5 in September 2025, Sonnet 4.6 in February 2026) suggested a gap of four to five months between point releases instead. And the version number 4.8 was no guarantee for the next public Sonnet model: Anthropic skipped it and released Sonnet 5 on 30 June 2026.

No specific features or capabilities. Other blogs speculated about "vision upgrades", "better coding benchmarks" or "higher image resolution" for Sonnet 4.8. These guesses derived solely from the fact that Opus 4.7 had received improvements in exactly these areas in April 2026 (up to 2,576 pixels on the long edge, 98.5 percent on customer XBOW’s visual-acuity benchmark according to Anthropic’s announcement, Task Budgets, file-system memory recall). There was not a single hint in the leak that these features would be carried over to the Sonnet tier, let alone in a version 4.8. That was reasoning by analogy, not evidence.

No pricing information. At the time of the leak, Sonnet 4.6 cost 3 US dollars per million input tokens and 15 US dollars per million output tokens. Some sources claimed Sonnet 4.8 would "probably" have the same price. That was not information derived from the leak but an extrapolated guess, and it did not hold: Sonnet 5 arrived at 2 and 10 US dollars, cheaper than Sonnet 4.6, and Sonnet 5.5 followed on 28 September 2026 at the same price.

No API model identifier. Anthropic documents its API models in a public list. There was never an entry for Sonnet 4.8. The leak likewise contained no API endpoint or internal routing code that references claude-sonnet-4-8 or a similar ID.

The leak hierarchy: what else was in the code

To place Sonnet 4.8 correctly, it is worth looking at the other finds from the March leak. They showed what Anthropic was working on in parallel and how far advanced these features were compared to a possible new Sonnet version.

KAIROS — the always-on background agent. With more than 150 mentions in the source code according to VentureBeat, KAIROS was the most advanced unreleased component. It was a persistent daemon that runs even when the developer is not at the machine. According to an analysis by Akita on Rails, KAIROS receives periodic <tick> prompts, decides autonomously whether it should act, and has exclusive tools such as push notifications, file delivery and GitHub webhook subscriptions. Proactive actions that would block the user for more than 15 seconds are deferred. The feature was fully implemented, but hidden behind compile-time feature flags. The architectural relevance is enormous: KAIROS would upgrade Claude Code from a reactive chat tool to a proactive, self-directed system partner. For Sonnet 4.8 there was no comparable evidence within the leak.

Capybara, an internal model codename. Which model it stands for is open: VentureBeat and InfoQ map Capybara to a Claude 4.6 variant, and Anthropic has not confirmed this. In Anthropic’s own draft blog post, which became public through a data leak shortly before the code leak, Capybara instead named a new model tier above Opus; a second version of the same draft used the name Mythos in its place (Fortune, The Decoder). According to VentureBeat, the leak contains internal quality metrics that are remarkably transparent for the industry: code comments put the false-claims rate of Capybara v4 at 16.7 percent and that of v8 at 29 to 30 percent. What exactly this internal measure of false claims counts is not public. The code also documents an "assertiveness counterweight" meant to prevent the model from rewriting user code too aggressively. This data is more valuable than any marketing benchmark, because it consists of genuine internal engineering notes.

Numbat — the unknown model. Alongside Capybara, another tier codename exists in the code: Numbat. One comment reads verbatim "Remove this section when we launch numbat". This proved that Numbat was a standalone model in the pre-launch phase, but there was no information on tier, capabilities or timeline.

ULTRAPLAN, Coordinator Mode, Buddy System. According to the analysis by Akita on Rails, the leak contains 44 compile-time feature flags. ULTRAPLAN offloads complex architecture planning into remote cloud containers with up to 30 minutes of compute time. Coordinator Mode enables multi-agent swarms in which one Claude instance can spawn and manage parallel worker agents. The Buddy System is a Tamagotchi-like ASCII companion in the terminal — apparently conceived as an Easter egg for April 1, 2026. All of these features were fully present in the code, but set to false for external builds.

Why the version number 4.8 instead of 4.7 confused

One of the most frequently asked questions in the forums was: why is there no Sonnet 4.7? Anthropic released Opus 4.7 on April 16, 2026. There is no fixed pattern by which Sonnet follows Opus: Opus 4 and Sonnet 4 shipped on the same day, Sonnet 4.5 eight weeks before Opus 4.5, and only with 4.6 did Sonnet follow Opus, twelve days later (release notes). Some analysts nonetheless speculated that a Sonnet 4.7 would appear shortly after Opus 4.7. Instead, only the string "sonnet-4-8" was found in the leak.

Several explanations circulated, all unproven: a jump of the Sonnet series from 4.6 to 4.8 to restore parity with the Opus version; an internal Sonnet 4.7 that was never released and merged into 4.8; or an arbitrary numbering that said nothing about the actual feature scope. In the end Anthropic jumped from Sonnet 4.6 straight to Sonnet 5, and the version number 4.8 never appeared publicly.

What developers and companies should take from it

From our perspective as an agency that advises mid-market companies on AI integration, three lessons can be derived that apply well beyond this leak.

First: do not make architecture decisions based on leaks. Anyone who waited for a Sonnet 4.8 in the spring waited for a model that never came. The right choice was the model that was available: until the end of June Sonnet 4.6, which users preferred over Sonnet 4.5 roughly 70 percent of the time in early Claude Code testing, according to Anthropic, then Sonnet 5 and, from 28 September 2026, Sonnet 5.5. Thanks to abstraction layers in the API integration, a model switch is usually a configuration change, not a rewrite. Leak-based pre-emptive migrations are technical debt without return.

Second: keep an eye on the agentic features. If the leak made one thing clear, it is that Anthropic is shifting its focus massively from pure model upgrades to agentic capabilities. In the leak, KAIROS, Coordinator Mode and ULTRAPLAN were not marginal experiments but fully implemented subsystems. For developers this means: the question will no longer be "Which model do I use?" but "How do I orchestrate multiple agents working autonomously?". Anyone optimizing their infrastructure for pure prompting is missing the real paradigm shift.

Third: train source literacy. Most articles about Sonnet 4.8 in the German- and English-speaking world reinforced one another without going back to primary sources. The original tweet that discovered the string itself carried a question-mark qualifier. Wavespeed.ai rated the corroboration as low (“low corroboration”), and rightly so. We recommend that every technology decision-maker check at least two independent primary sources before strategic decisions. For this leak those are the GitHub mirror of the original source maps (largely removed through DMCA takedowns) and the detailed analyses by DecodeTheFuture.org and Kuber Studio, which documented the original code promptly.

Conclusion

Claude Sonnet 4.8 existed as an internal version string at Anthropic, and that is all the Claude Code leak from March 31, 2026 ever proved about it. The model never shipped; after Sonnet 4.6 came Sonnet 5. Every statement about features, prices, release dates or benchmarks was speculation, partly plausible, but never derived from the leak. The true value of the leak lay elsewhere: it showed that Anthropic already had models, agent frameworks and safety mechanisms in March that could push the public perception of "AI assistance" forward by a decade.

For companies the consequence is clear: invest in robust, model-agnostic architectures and in understanding agentic workflows, not in waiting for the next version number. Sonnet 5.5 has been available since 28 September 2026; KAIROS and Mythos show where things are heading. Whoever can tell the difference wastes no budget on rumors.

Frequently asked questions about Claude Sonnet 4.8