Apex

Callstack introduced Apex on 5/27/2026 as a fast, specialized coding model aimed at React Native development. The article frames Apex as a focused alternative to general-purpose coding models for mobile app work. The post appears within Callstack's AI article series and references surrounding tooling topics. The excerpt positions Apex as a purpose-built model rather than a general frontier model, tailored to React Native code generation. It predates the 10/1/2026 GA announcement and functions as historical grounding for the model's origins. Exact benchmark results, parameter counts, or context window details are not visible in the supplied text.

Callstack announced that Apex is now generally available as a coding model specialized for React Native, Next.js, and web development. The model is positioned for agentic coding workflows across mobile and web. The post also references related tooling and ecosystem updates from the Callstack blog. Callstack describes Apex as a custom coding model for React and React Native projects, with an emphasis on agentic, multi-step coding tasks. The announcement is dated 10/1/2026 and appears on callstack.com alongside other AI engineering articles. Specific benchmark numbers, API pricing, and self-hosting details are not visible in the supplied excerpt.

The LLM Reference per-model entry for Apex confirms it was released 2026-10-01 as a Proprietary coding model created by Callstack, the Wrocław, Poland-based React and React Native core contributor founded in 2016. Specifications include 262,144-token context, Reasoning and JSON/Tool use support, OpenAI- and Anthropic-compatible endpoints enabling integration with Claude Code, Cursor and Codex, and $0.500 per 1M input / $3.00 per 1M output pricing with a $0.200 cache read rate, identical across Callstack Apex API and Vercel AI Gateway. According to the same reference page, Apex is described as Callstack's agentic coding model for mobile and web development, post-trained on React and React Native source code, widely used open-source React Native libraries, and Callstack's own engineering documentation, making it strongest for React Native, React and Next.js work including feature building, debugging, code review, tests, upgrades and migrations to the New Architecture. The model is explicitly less capable outside that domain than general-purpose models of similar size, and serves only as an inference route through Vercel AI Gateway.

Callstack's Apex is officially documented as a Qwen-based language model specialized for React Native and React engineering, available via the `apex` alias or the pinned snapshot `apex-20261001`, both generally available since 1 October 2026. The publisher page lists a 262,144-token context window, text input/output, and a base model from Qwen (Alibaba Group) used under licence, then post-trained on React and React Native source, ecosystem libraries, and Callstack's internal documentation. According to the apex.callstack.com models page, training and processing occur in the European Economic Area with zero data retention always on, meaning prompts and outputs are never stored for any traffic. The model targets greenfield and brownfield development, native module integration, New Architecture adoption, upgrades and migrations, while being explicitly limited outside React Native and React domains and capable of producing incorrect code that requires review.

Callstack's Apex coding model, specialized for React Native with strong Next.js performance, reached general availability on October 1, 2026, after a stealth preview on Vercel AI Gateway under the name Pixel Canary. It is built on a Qwen foundation with Callstack's own domain training pipeline and is aimed at agentic coding workflows including debugging, review, and testing. Dedicated iOS and Android models are planned next, and commercial self-hosting is available alongside the API. API pricing starts at $0.50 per million input tokens, $0.20 for cached input, and $3.00 per million output tokens, which the post frames as roughly 70% less than GPT-6 Sol's output pricing and 85% less than Claude Opus 5.5's. In a recorded code review on the Expensify React Native repository, Apex completed the task in 2 minutes 16 seconds for $0.09, while Claude Opus 5.5 took 13 minutes 30 seconds for $0.60, about 6x faster and 85% cheaper on that single run.

Apex is Callstack's agentic coding model optimized for React Native on mobile and Next.js on the web, positioned as a cost-effective general-purpose model for everyday agent work such as debugging, code review, testing, and large refactors. The model ships with a 256K context window for longer repository context, plugs into existing tooling like Claude Code, Cursor, and Codex via an endpoint swap, and is reported to run at up to 250 tokens per second of throughput, roughly 3x faster than GPT 6.1 Sol. On published benchmarks, Apex scores within about 1 point of Claude Opus 5.5 on React Native evals (87.8% vs 88.6%) and matches Claude Opus 5.5 on Next.js evals (96.8% pass rate). Per-task cost is $7.72 on React Native versus $28.75 for Claude Opus 5.5 and $0.067 on Next.js versus $0.891 for GPT 6 Astra. API list pricing is $0.50 per 1M input tokens, $3.00 per 1M output tokens, with cached input billed at 10% of the input rate, per the official Apex page.