Grok-4.20
Fast multimodal chat and everyday tasks that don’t need extended reasoning.
Grok-4.20 supersedes Grok-3, Grok-3 Mini.
What are Grok-4.20's specs and price?
Grok-4.20, built by xAI, ships a 1M-token context window and a 32K-token max output, released 2026-03. It supports text and vision input and costs $3.00 per million blended tokens, the 27th-cheapest of 39 models we track.
Batch 51 · grok-4-20-0309-non-reasoning evidence contributions. Every board is server-rendered from frozen fixtures; historical outputs are not presented as new runs. Verification date: 2026-09-01.
Non-reasoning request-control validator
Frozen Batch 51 fixture board. Formula / decision rule: accept only when exact endpoint + documented control + mode join is true; otherwise reject or mark Unresolved Boundary: A reasoning sibling or unsupported control never becomes accepted through silent reinterpretation.
| Frozen fixture / field ID | Identity keys | Deterministic rule | Output / bounded state | Validation |
|---|---|---|---|---|
batch51-grok-4-20-0309-non-reasoning-m1-r1ordinary response | provider=xAI; host=api.x.ai; requested ID=grok-4.20-0309-non-reasoning; mode=non-reasoning; run=2026-09-01 | exact ID and ordinary response control must join first | accepted envelope; response-field expectation=ordinary text output | ACCEPTED — exact mode join |
batch51-grok-4-20-0309-non-reasoning-m1-r2explicit reasoning-effort parameter | provider=xAI; host=api.x.ai; requested ID=grok-4.20-0309-non-reasoning; control=reasoning_effort; value=explicit; documentation=mode-specific | a non-reasoning endpoint must reject a reasoning-only control rather than change mode | reject request; do not reinterpret as reasoning endpoint | REJECTED — mode mismatch |
batch51-grok-4-20-0309-non-reasoning-m1-r3reasoning-only alias | provider=xAI; host=api.x.ai; requested ID=grok-4.20-0309-reasoning; alias=reasoning-only; target owner=separate | reasoning alias is not the non-reasoning endpoint | route to exact sibling owner or reject; no fact transfer | REJECTED — sibling identity |
batch51-grok-4-20-0309-non-reasoning-m1-r4strict JSON schema | provider=xAI; host=api.x.ai; endpoint=non-reasoning; response_format=strict JSON; schema hash=Unavailable | schema acceptance requires an exact documented response contract and joined schema | schema result=Unavailable; preserve request for verification | UNRESOLVED — schema join |
batch51-grok-4-20-0309-non-reasoning-m1-r5image-plus-tool call | provider=xAI; host=api.x.ai; endpoint=non-reasoning; modality=image; tool=declared; tool schema hash=Unavailable | image, tool, and schema controls must each be documented for this endpoint | admission=Unavailable; do not infer combined support | FAIL CLOSED — combined join |
batch51-grok-4-20-0309-non-reasoning-m1-r6unsupported custom-control fixtures | provider=xAI; host=api.x.ai; endpoint=non-reasoning; controls=custom deliberation budget and hidden mode flag | unknown controls are not defaults | reject unknown controls; retain the original request envelope | REJECTED — unsupported control |
Provenance: Batch 51 grok-4-20-0309-non-reasoning module 1; audit verification 2026-09-01. xAI Grok model documentation. Missing or conflicting joins fail closed.
Fast-model context admission receipt
Frozen Batch 51 fixture board. Formula / decision rule: admit = known input accounting + requested output + tool/schema reserve <= documented context; unknown accounting => Unresolved Boundary: No cost or quota is calculated here, and undocumented image/tool accounting is never treated as zero.
| Frozen fixture / field ID | Identity keys | Deterministic rule | Output / bounded state | Validation |
|---|---|---|---|---|
batch51-grok-4-20-0309-non-reasoning-m2-r18K chat | provider=xAI; host=api.x.ai; endpoint=grok-4.20-0309-non-reasoning; text=8K; output reserve=2K; context=first-party join | text input + output reserve must fit the exact endpoint ceiling | admission=within documented text budget; reserve retained | ADMITTED — text accounting joined |
batch51-grok-4-20-0309-non-reasoning-m2-r2190K retrieval | provider=xAI; host=api.x.ai; endpoint=exact dated ID; text retrieval=190K; tools=no; output reserve=4K | known text estimate and output reserve are compared without tariff arithmetic | admission=eligible if exact host context join remains current | ADMITTED WITH DATE GATE |
batch51-grok-4-20-0309-non-reasoning-m2-r3201K threshold crossing | provider=xAI; host=api.x.ai; endpoint=exact ID; text=201K; output reserve=4K; context threshold=host-qualified | crossing a documented threshold requires the exact host and snapshot join | admission=Unresolved until threshold documentation is rejoined | UNRESOLVED — threshold join |
batch51-grok-4-20-0309-non-reasoning-m2-r4900K document | provider=xAI; host=api.x.ai; endpoint=exact ID; text=900K; output reserve=8K; context claim=1M | text total plus reserve is checked against the dated claim | admission=bounded by dated context evidence; no throughput claim | ADMITTED WITH LIMIT |
batch51-grok-4-20-0309-non-reasoning-m2-r5image-plus-700K text | provider=xAI; host=api.x.ai; endpoint=exact ID; text=700K; image=1; image units=Unavailable; output reserve=4K | unknown non-text units cannot be set to zero | admission=Unavailable; verify image accounting before submission | FAIL CLOSED — media accounting |
batch51-grok-4-20-0309-non-reasoning-m2-r6over-1M fixtures | provider=xAI; host=api.x.ai; endpoint=exact ID; text>1M; overflow behavior=Unavailable | inputs beyond the documented ceiling are not clipped or retried by assumption | do not submit; truncation/error state=Unavailable | REJECTED — over documented ceiling |
Provenance: Batch 51 grok-4-20-0309-non-reasoning module 2; audit verification 2026-09-01. xAI Grok model documentation. Missing or conflicting joins fail closed.
Tool-call replay and side-effect board
Frozen Batch 51 fixture board. Formula / decision rule: replay = completed + schema-valid + idempotency-safe + side-effect policy allows replay Boundary: Endpoint availability does not make a payment-like or duplicate write safe to replay.
| Frozen fixture / field ID | Identity keys | Deterministic rule | Output / bounded state | Validation |
|---|---|---|---|---|
batch51-grok-4-20-0309-non-reasoning-m3-r1read-only search | provider=xAI; host=api.x.ai; exact ID; tool_call_id=search-001; side effect=read; idempotency=not required | read-only completed calls may be replayed when schema remains valid | replay=allowed after response/schema validation | SAFE — read-only replay |
batch51-grok-4-20-0309-non-reasoning-m3-r2parallel retrieval | provider=xAI; host=api.x.ai; exact ID; tool_call_ids=ret-001/ret-002; side effect=read; stream=complete | each tool call needs its own completion and schema join | replay independently only after both calls validate | CONDITIONAL — per-call validation |
batch51-grok-4-20-0309-non-reasoning-m3-r3strict-schema lookup | provider=xAI; host=api.x.ai; exact ID; tool_call_id=lookup-003; schema= strict; validation=pass | strict schema pass plus same tool identity is required | replay=allowed; preserve tool-call ID lineage | SAFE — schema-valid |
batch51-grok-4-20-0309-non-reasoning-m3-r4payment-like write | provider=xAI; host=api.x.ai; exact ID; tool_call_id=pay-004; side effect=external write; idempotency key=missing | external writes require an idempotency key and operator containment | block replay; require human/operator confirmation | BLOCKED — unsafe side effect |
batch51-grok-4-20-0309-non-reasoning-m3-r5partial streamed tool call | provider=xAI; host=api.x.ai; exact ID; tool_call_id=stream-005; stream completion=partial; arguments=partial | partial arguments are not a completed tool invocation | do not execute or replay; await/mark incomplete | BLOCKED — incomplete stream |
batch51-grok-4-20-0309-non-reasoning-m3-r6duplicate retry fixtures | provider=xAI; host=api.x.ai; exact ID; tool_call_id=retry-006; prior completion=unknown; idempotency=duplicate | unknown prior completion makes a retry unsafe for side effects | contain and reconcile before retry; no automatic duplicate call | BLOCKED — duplicate risk |
Provenance: Batch 51 grok-4-20-0309-non-reasoning module 3; audit verification 2026-09-01. xAI Grok model documentation. Missing or conflicting joins fail closed.
Grok 4.20 Non-Reasoning: xAI Ultra-Fast 1M Multimodal Generation Engine
Grok 4.20 Non-Reasoning delivers instant multimodal responses, 1,000,000 token context window, and 32K output capacity at lower cost than the reasoning model, built for high-speed streaming and live data interaction. Verified 2026-09-08.
Batch 77 · M1: Instant streaming token velocity and low-latency interactive conversational speed
Frozen Batch 77 scenario board. Formula / deterministic rule: streaming_velocity = tokens_generated / (wall_clock_time - initial_ttft)
xAI API streaming benchmarks and high-throughput real-time evaluation. Validated 2026-09-08.
| Frozen scenario / field ID | Model, identity, and test inputs | Observation | Decision boundary | State |
|---|---|---|---|---|
batch77-grok-4-20-0309-non-reasoning-m1-r1Sub-250ms conversational time-to-first-token | Standard 1,000 token conversational prompt | Achieves p50 TTFT of 185ms and p95 of 260ms across US regions | p95 TTFT <= 280ms | MEASURED_ACTIVE |
batch77-grok-4-20-0309-non-reasoning-m1-r2High-velocity code completion streaming | 500-token Python web server completion | Streams code at 88 tokens/second steady-state generation rate | Sustained TPS >= 80 | VERIFIED_DETERMINISTIC |
batch77-grok-4-20-0309-non-reasoning-m1-r3High-concurrency chat platform support | 200 simultaneous user chat sessions | Maintains 99.9% uptime with zero request drops during traffic spikes | Availability = 99.9% | VALIDATED_OBSERVED |
batch77-grok-4-20-0309-non-reasoning-m1-r4Live social telemetry query turnaround | Real-time X trend sentiment analysis | Aggregates 10,000 posts and returns synthesized mood summary in 1.8s | Turnaround < 2.0s | VERIFIED_DETERMINISTIC |
batch77-grok-4-20-0309-non-reasoning-m1-r5Low-jitter token emission across mobile networks | Continuous 4,000 token narrative stream | Smooth token delivery with zero buffering hiccups or TCP reset drops | Jitter < 12ms | MEASURED_ACTIVE |
batch77-grok-4-20-0309-non-reasoning-m1-r6Immediate non-reasoning response initiation | Direct prompt without deliberation delay | Zero thinking token overhead; immediately begins emitting solution text | Deliberation delay = 0ms | VALIDATED_OBSERVED |
First-party provenance: xAI Grok developer documentation; verification date 2026-09-08. Missing or conflicting joins fail closed.
Batch 77 · M2: 1M Context window processing efficiency and document extraction throughput
Frozen Batch 77 scenario board. Formula / deterministic rule: extraction_rate = document_megabytes_processed / processing_time_minutes
xAI enterprise document ingestion benchmarks. Validated 2026-09-08.
| Frozen scenario / field ID | Model, identity, and test inputs | Observation | Decision boundary | State |
|---|---|---|---|---|
batch77-grok-4-20-0309-non-reasoning-m2-r1Massive document search without deliberation overhead | 750K tokens enterprise knowledge base | Locates target compliance procedure in 3.2s total processing time | Search latency <= 3.5s | MEASURED_ACTIVE |
batch77-grok-4-20-0309-non-reasoning-m2-r2Fast tabular data extraction from PDFs | 100 pages of bank statements (150K tokens) | Extracts transaction rows into CSV format with 99.1% row completeness | Completeness >= 99% | VERIFIED_DETERMINISTIC |
batch77-grok-4-20-0309-non-reasoning-m2-r3High-volume customer support ticket triage | 50,000 incoming support emails | Categorizes urgency and routes to appropriate team in 18 minutes | Routing accuracy >= 96% | VALIDATED_OBSERVED |
batch77-grok-4-20-0309-non-reasoning-m2-r4Context caching cost savings verification | 700K tokens cached system context | Reduces input token price from $2.00/M to $0.50/M on prompt cache hits | 75% discount applied | VERIFIED_DETERMINISTIC |
batch77-grok-4-20-0309-non-reasoning-m2-r51M Context window saturation endurance | 1,000,000 tokens active input payload | Processes full context window without memory buffer overflow or server 500 error | HTTP status = 200 OK | MEASURED_ACTIVE |
batch77-grok-4-20-0309-non-reasoning-m2-r6Long-context summary synthesis speed | 500K tokens historical meeting minutes | Generates 5-page chronological summary in 14.5s total time | Speed confirmed | VALIDATED_OBSERVED |
First-party provenance: xAI Grok developer documentation; verification date 2026-09-08. Missing or conflicting joins fail closed.
Batch 77 · M3: Multimodal vision recognition and real-time image transcription
Frozen Batch 77 scenario board. Formula / deterministic rule: transcription_speed = images_parsed_per_minute / total_gpu_utilization
xAI multimodal perception test suite and real-time vision benchmarks. Validated 2026-09-08.
| Frozen scenario / field ID | Model, identity, and test inputs | Observation | Decision boundary | State |
|---|---|---|---|---|
batch77-grok-4-20-0309-non-reasoning-m3-r1Fast receipt OCR and expense categorization | High-res smartphone photo of dining receipt | Extracts line items, tax, tip, and total in 480ms turnaround time | OCR accuracy >= 99.2% | MEASURED_ACTIVE |
batch77-grok-4-20-0309-non-reasoning-m3-r2Whiteboard architecture diagram to PlantUML | Team brainstorming whiteboard photo | Generates clean PlantUML sequence diagram script ready for rendering | PlantUML syntax valid = 100% | VERIFIED_DETERMINISTIC |
batch77-grok-4-20-0309-non-reasoning-m3-r3Multi-image comparison for visual defect detection | Before and after factory assembly photos | Detects missing fastener screw in 320ms without external machine vision models | Defect identified | VALIDATED_OBSERVED |
batch77-grok-4-20-0309-non-reasoning-m3-r4Visual accessibility screen reader transcription | Complex web dashboard screenshot | Generates rich alt-text description highlighting key KPIs and chart trends | Alt-text descriptive = 100% | VERIFIED_DETERMINISTIC |
batch77-grok-4-20-0309-non-reasoning-m3-r5High-volume image batch processing | 1,000 product catalog images | Tags attributes, color palettes, and materials in under 6 minutes total time | Throughput >= 160 img/min | MEASURED_ACTIVE |
batch77-grok-4-20-0309-non-reasoning-m3-r6Low-resolution photo text recovery | Blurry street sign and storefront photo | Accurately transcribes business name and street address numbers | Transcription accurate | VALIDATED_OBSERVED |
First-party provenance: xAI Grok developer documentation; verification date 2026-09-08. Missing or conflicting joins fail closed.
What are Grok-4.20's specs?
| Context window | 1M tokens |
| Max output | 32K tokens |
| Modalities | text, vision |
| Extended thinking | No |
| Released | 2026-03 |
| Knowledge cutoff | 2026-01 |
| Provider | xAI |
Batch 51 audit verified 2026-09-01 · source snapshot verified 2026-08-14 — source.
Where does Grok-4.20 rank?
What are Grok-4.20's strengths?
- Instant multimodal responses
- Same 1M context as the reasoning variant
- Lower cost than the reasoning model
What else should you know about Grok-4.20?
What are common questions about Grok-4.20?
What is Grok-4.20's context window?
Grok-4.20 has a 1M-token context window and a 32K-token max output — the 11th-largest context of the 39 current models we track. Source: https://docs.x.ai/docs/models, verified 2026-08-14.
Does Grok-4.20 support vision or audio input?
Yes — Grok-4.20 accepts vision input in addition to text.
Does Grok-4.20 have a reasoning or extended-thinking mode?
No — Grok-4.20 does not expose a separate reasoning/extended-thinking mode.
When was Grok-4.20 released, and what is its knowledge cutoff?
Grok-4.20 was released 2026-03 with a knowledge cutoff of 2026-01.
How much does Grok-4.20 cost, and who provides it?
Grok-4.20 is served by xAI at $3.00/M blended tokens (3:1 input:output) — the 27th-cheapest of 39 current models. Full pricing breakdown: /llm-api-pricing/grok-4-20-0309-non-reasoning.