Claude Sonnet 5 Rumored to Be Faster and Cheaper, Leaks Claim

While OpenAI continues removing older ChatGPT models, fresh leaks suggest Anthropic may be preparing its next major release. Weekend reports point to a new model called Claude Sonnet 5, hinting at a significant leap in performance and pricing strategy.

According to TestingCatalog, leaked references show a build labeled Claude Sonnet 5 with an internal date string of February 3, 2026. It remains unclear whether this date signals a public launch or an internal milestone, but the timing would place it just days before Super Bowl LX on February 8.

Claude Sonnet 5, reportedly codenamed “Fennec”

Information shared on X claims that Claude Sonnet 5 carries the internal codename Fennec. Leaks describe it as a full-generation step ahead of Google Gemini, internally referred to as “Snow Bunny.” This aligns with a broader trend of AI labs pushing harder on mainstream visibility to compete with ChatGPT and Gemini.

TestingCatalog reports that early hands-on testing shows impressive math capabilities for the non-thinking Sonnet 5 variant. In some workflows, Sonnet 5 reportedly outperforms Claude Opus 4.5 in coding tasks.

Testers also highlighted structured visual generation as a standout strength. One example involved an ASCII world map prompt that produced unusually complete and detailed output. UI-focused and rendering-related coding tasks reportedly showed similarly strong results.

Context window, pricing, and infrastructure rumors

The observed test build lists a 128k context window, though other sources claim the final version could retain a massive 1 million token context window while running significantly faster than previous Sonnet models.

Pricing rumors suggest an aggressive approach. Sonnet 5 is said to cost around 50% less than Claude Opus 4.5, while delivering stronger overall performance. Insider leaks also claim scores above 80.9% on SWE-Bench, potentially putting it ahead of current top coding models.

Sonnet 5 reportedly targets a faster, lower-cost tier and may be optimized for Google TPUs, enabling higher throughput and reduced latency. A separate clue points to Google infrastructure readiness, as a 404 error on a specific Sonnet 5 Vertex ID suggests the model already exists internally and awaits activation.

If the rumored pricing, latency, and benchmarks hold, Claude Sonnet 5 could emerge as a general-purpose workhorse model for developers and everyday users. This direction fits Anthropic’s recent strategy of offering broadly usable, cost-efficient models rather than focusing only on premium tiers.

For now, all details remain unconfirmed, and timelines, pricing, and performance claims should be treated with caution.

In other AI news, OpenAI and Snowflake have announced a new partnership, and OpenAI has also launched its Codex app on macOS.