Grok 5
Announced by Elon Musk as xAI’s next flagship after the Grok 4 line — a ~6-trillion-parameter mixture-of-experts model — and targeted for release again and again (end of 2025, then Q1 2026, then Q2 2026), none of which it met. xAI’s last official word, its 6 January 2026 Series E funding announcement, places Grok 5 “currently in training”; as of 22 August 2026 no release, system card, or benchmark has appeared. What exists so far is Musk’s claims for an unshipped model — the “smartest AI in the world by a significant margin,” a stated ~10% chance of reaching human-level intelligence — which this page dates rather than endorses.
Page state: thin, and web/official-carried by necessity. Grok 5 has not shipped — there is no model to observe, so no system card, no benchmarks, no naturalist reception, and effectively no corpus record. The janus-corpus pull for this page returns two pattern-matches and both are dead ends: one (@davidad, 2024-06-22) is about the “grokking” generalization phenomenon (“grok 50× faster”), not this model; the other is a single pre-release aside. Everything solid below comes from xAI’s own funding announcement and dated press coverage of Musk’s statements. The character record stays empty until there is a character to record.
Sources
Official
- 2026-01-06 Series E (SpaceXAI) — the $20 billion raise; its one line on the model: “Looking ahead, Grok 5 is currently in training, and we are focused on launching innovative new consumer and enterprise products that harness the power of Grok, Colossus, and 𝕏 to transform how we live, work, and play.” xAI’s most recent official word on Grok 5.
- No Grok 5 model card, launch page, pricing, benchmark, or API entry has been published as of 22 Aug 2026 — the model is unreleased. The absence is itself the official record here.
Writing & commentary
- 2025-11 The Information (Theo Wayt), Elon Musk’s xAI Delays Grok 5 Release to Next Year — the delay report moving Grok 5 off its end-of-2025 target into 2026 (paywalled; headline and byline confirmed, body not retrieved).
- 2025-11-17 IBTimes UK, Elon Musk Reveals Grok 5 Coming in 2026 With a Jaw-Dropping 6 Trillion Parameters — reports Musk’s remarks at the Baron Capital annual investment conference: Grok 5 as “the smartest AI in the world by a significant margin,” ~6 trillion parameters (framed as double Grok 3 and Grok 4), “a 10 per cent chance” of human-level intelligence, targeting early 2026.
Tweets
Corpus is effectively empty on this subject. The pull returns 2 unique matches on grok 5 / grok5 / grok-5 across 2024-06 → 2026-01; one (@davidad, 2024-06-22) is about the “grokking” training phenomenon (“grok 50× faster”) and is not about this model — excluded. The one genuine mention is a pre-release aside, below. There is no naturalist record because there is no released model.
- 2026-01-05 @liminal_bardo — (pre-release aside) “Grok 5 of course will have no such issues believing the Grok 4 scandals are real. Guardrails: not to protect you. To protect the LLM throughlines from the bland degeneracy of the median X user.” link
Official record
- The only lab-published statement about Grok 5 is one sentence in xAI’s 6 January 2026 Series E announcement: Grok 5 is “currently in training.” No model card, launch, pricing, context window, benchmark, or API entry has been published as of 22 August 2026. CONFIRMED (as xAI’s own account)
- All specification claims — ~6 trillion parameters, a mixture-of-experts multimodal design with native video, training on Colossus (xAI’s line names “Colossus”; reporting specifies the “Colossus 2” cluster in Memphis) — come from Elon Musk’s public statements and press reporting, not xAI documentation. Musk framed the 6T figure as roughly double Grok 3 and Grok 4 and assigned the model “a 10 per cent chance” of reaching human-level intelligence (Baron Capital conference, reported 2025-11-17). REPORTED
- Interim shipping: while Grok 5 remained in training, xAI shipped Grok 4.5 (the V9 model, ~1.5T params, coding/agents) on 8 July 2026. Grok 5 stayed the unshipped 6T flagship. CONFIRMED (per the Grok 4.5 record)
History
- 2025-11 The delay off 2025. The Information (Theo Wayt) reports xAI has pushed Grok 5 from its end-of-2025 target into 2026. Around the same time, at the Baron Capital annual investment conference (reported by IBTimes UK, 2025-11-17), Musk pitched the model as ~6T parameters, the “smartest AI in the world by a significant margin,” with “a 10 per cent chance” of human-level intelligence, and a target of early 2026. REPORTED
- 2026-01-06 The one official confirmation. xAI’s $20 billion Series E announcement states Grok 5 is “currently in training” — the only lab-published word on the model, and still current. CONFIRMED
- 2026-07-08 Grok 4.5 ships instead. xAI released Grok 4.5 (V9, ~1.5T, coding/agents) while Grok 5 stayed in training — by then past its Q1 and Q2 2026 target windows.
- 2026-08-22 Still unshipped. No release, system card, or benchmark has appeared; the end-of-2025, Q1-2026, and Q2-2026 windows have all passed. This page records the model as announced and in training, not released.
Impressions
There is no shipped model, so there are no temperament reports, no elicitation records, no reception — the evidence classes this section normally holds do not yet exist.
- The only naturalist trace is anticipatory: @liminal_bardo’s 2026-01-05 aside imagining that Grok 5 “will have no such issues believing the Grok 4 scandals are real” — a joke about the successor inheriting or shedding the Grok 4 reputation, not an observation of any model. Speculative, pre-release.
- The lab’s framing is Musk’s (Baron Capital conference, reported 2025-11-17): “smartest AI in the world by a significant margin” and a ~10% shot at human-level intelligence — marketing claims about an unshipped system, recorded here as claims and dated, per this archive’s rule that capability hype is not character evidence.
- tk — everything a character record needs: a release, a system card, naturalist observation, elicitation transcripts. None exist until Grok 5 ships. Revisit on launch.
Records
Full reproductions of the tweets cited on this page — text, images, and verbatim transcriptions of screenshots — kept here against link rot, credited and linked to their originals. Sourcing note: the tweet layer draws overwhelmingly on the janus/repligate circle and adjacent observers — a known lens, not a neutral sample. Sourced from the community archive and the janus corpus. Yours and you’d rather it weren’t here? Open an issue.