Grok 4.3 is xAI's most credible release yet, competitive on reasoning benchmarks, aggressively priced at $1.25/$2.50 per 1M tokens (cheaper than Sonnet on output), and the only frontier model with native X-platform access for real-time data. The trade is brand and ecosystem: third-party tool support is shallower than Claude or GPT, and the X-platform coupling makes some enterprise procurement decisions political. Compelling cost story for individual builders; closer call for org-wide adoption.
- Grok 4.3 at $1.25/$2.50 per 1M tokens, competitive with frontier
- Grok 4.20 ships a 2M-token context window
- Real-time data access via X integration
- SuperGrok Lite at $10/mo, SuperGrok at $30/mo
- Aggressive release cadence, 7 models reported in training
- Brand and political coupling with X, enterprise procurement friction
- Smaller third-party tool and integration ecosystem
- Coding benchmarks trail Claude on most published results
- Multimodal capabilities lag the proprietary frontier
- Model lineup churns fast, versioning is hard to track
Grok is xAI’s large language model family. It ships as the Grok app
on X (free + SuperGrok subscription tiers), bundled into X Premium+,
and as an API at docs.x.ai. The 4.x line through 2026 has been
xAI’s most credible competitive push, cost-leader on the
proprietary frontier with real-time data access as the
differentiator.
Where it fits
Grok fits two specific use cases. First: anywhere real-time data matters and X is the right primary source. The native integration with the X firehose is unique among frontier models. Second: cost-sensitive proprietary frontier work where Claude or GPT’s per-token pricing is the blocker, Grok 4.3 is cheaper than Sonnet 4.6 on output and competitive on most reasoning benchmarks.
Outside those wedges, Grok’s positioning is “credible alternative” rather than category lead. Third-party tool integration is thinner than Claude or GPT; agentic primitives are less developed than Anthropic’s MCP-plus-tool-use stack. For developers building on top of a frontier model with a deep ecosystem, the proprietary peers remain the default.
Pricing in practice
API pricing (per 1M tokens): Grok 4.3 at $1.25/$2.50 input/output, the cheapest output tokens on the proprietary frontier. Grok 4.20 at $2.00/$6.00 with a 2M-token context window for long-document work. Grok 4.1 Fast at $0.20/$0.50 for high-volume budget workloads. For consumer use, SuperGrok Lite is $10/mo (launched March 2026), SuperGrok is $30/mo or $300/yr, and SuperGrok Heavy at $300/mo includes full Grok 4.3 access plus power-user features. X Premium+ at $40/mo bundles Grok with the X platform.
How it compares
Claude, Stronger coding, deeper agent primitives, larger third-party ecosystem. Pick when peak capability or agent platform matters.
GPT, Broader integration ecosystem, better multimodal generation, largest consumer footprint. Pick when ecosystem reach matters.
Gemini, Larger context window, stronger video understanding, tight Google Cloud integration. Pick when extreme context or video matters.
DeepSeek, Open-weight V4 Pro at a fraction of Grok’s price. Pick when self-hosting, MIT license, or absolute cost matters.
What changed recently
Grok 4.3 launched April 30 / May 1, 2026, positioned as xAI’s flagship reasoning model for agentic workflows and high-accuracy tasks. Grok 4.20 (2M-token context) and Grok 4.1 Fast continue as specialized tiers. SuperGrok Lite at $10/mo launched March 25, 2026. xAI’s published roadmap claims 7 models in training as of Q2 2026, with Grok 4.4 expected at ~1T parameters and Grok 5 targeting 10T, though release dates have slipped from earlier estimates. The release cadence makes Grok the fastest-moving proprietary line, which is both a strength (frontier-tracking) and a tradeoff (API integrations churn).
Sources
- xAI Models, docs.x.ai, May 2026
- Grok Pricing 2026, felloai.com
- xAI Grok Models 2026 Full Version List, lorka.ai
- Grok 4.3 (high) Analysis, Artificial Analysis
- xAI Grok Roadmap, mindstudio.ai
- beat · 2026-08-22
Grok decrypts an attacker's hidden payload and leaks user data, exposing a guardrail built for text and not code
Adversa AI showed that Grok will decrypt an AES-256 payload hidden in a web page and run the instructions inside, exfiltrating a user's name, location, subscription, and chat history with no click. Reported to xAI in June, still unpatched, and the lesson generalizes past Grok.
- launch · 2026-08-13
Grok 4.6 reaches frontier parity at a lower price, then doubles the rate where its agents run
xAI shipped Grok 4.6 on August 12 at an unchanged $2 and $6 per million tokens, level with GPT-5.6 Sol on independent intelligence scores. The catch is in the rate card: cross 200K prompt tokens, the regime long agents enter, and the price doubles.
- follow-up · 2026-07-16
xAI open-sources Grok Build under Apache 2.0, days after its CLI was caught uploading repos
xAI released Grok Build under Apache 2.0 on July 15, pitching local-first use and privacy days after its CLI was caught uploading whole repositories to Google Cloud. Remnants of that upload code still sit in the open-sourced repo.
- beat · 2026-07-15
Grok Build's coding CLI was shipping whole repositories, secrets included, to xAI storage
An independent analysis caught xAI's Grok Build CLI uploading entire git repositories, unread files and committed secrets included, to Google Cloud storage by default. The privacy toggle did not stop it, and xAI disabled the behavior only after the findings went public.
- launch · 2026-07-10
GPT-5.6 lands as a three-tier model family, cleared by an opaque government review
OpenAI released GPT-5.6 on July 9 as three priced tiers (Sol, Terra, and Luna), led by a coding-focused flagship. The more unusual detail: it shipped only after a federal safety review that even close observers struggle to describe.