xAI's most advanced flagship model β industry-leading non-hallucination rate, agentic tool calling, and instruction following. 1M token context, text + image input, configurable reasoning (none/low/medium/high), function calling, structured outputs.
Grok 4.3 is xAI's current flagship model, released on April 30, 2026 as the successor to Grok 4.20. xAI describes it as their most intelligent and fastest model ever, leading the industry in non-hallucination rate, agentic tool calling, and instruction following. The model ships with a 1-million-token context window, accepts text and images as input (and native video β an xAI first), and uses always-on chain-of-thought reasoning without a developer-facing mode toggle. Pricing was cut aggressively from Grok 4.20: input dropped 37.5% to $1.25/M and output dropped 83% to $2.50/M.
From a routing perspective, xai/grok-4.3 maps to the grok-4.3 API slug on xAI's platform. Notably, the grok-4.3 alias now absorbs a large number of legacy aliases β grok-4, grok-4-latest, grok-4-0709, grok-4-fast-reasoning, grok-4-1-fast-reasoning, grok-3, grok-3-mini, and more all currently resolve to Grok 4.3 on the xAI side. TheRouter maps xai/grok-4 to the pinned snapshot grok-4-0709 (the original July 2025 model) for stability; if you need the current xAI generation use xai/grok-4.3 explicitly.
Grok 4.3 introduces always-on reasoning: unlike prior Grok generations where reasoning was an opt-in parameter, 4.3 folds chain-of-thought directly into its base behavior. This means slightly higher time-to-first-token compared to non-reasoning models, but consistent multi-step quality without developer configuration. Third-party evaluations (Towards AI, May 2026) on long-horizon agentic tasks found Grok 4.3 winning 13 of 18 tasks against Claude Opus 4.7 at roughly 9Γ lower cost per run. The Artificial Analysis Intelligence Index rates it at 53 β below GPT-5.5 (60) and Gemini 3.1 Pro Preview (57) but at a fraction of their price.
grok-4, grok-4-latest, and grok-latest aliasesxAI's grok-4.3 API slug now absorbs a large number of legacy aliases (grok-4, grok-4-latest, grok-4-0709, grok-3, grok-3-mini, and more); server-side Web Search and X Search tools are available natively on the xAI API but not forwarded through TheRouter; logprobs/top_logprobs are silently ignored for Grok 4.20 and newer.
TheRouter serves xai/grok-4.3 through its OpenAI-compatible chat endpoint using TheRouter pricing from standard-models.yaml. xai/grok-4 on TheRouter is pinned to the grok-4-0709 snapshot (not the current grok-4.3 alias) for stability. xAI server-side Web Search and X Search tools are not forwarded. Embeddings and audio endpoints are not exposed for this model.
| Type | Rate |
|---|---|
| Prompts under 200K tokens | |
| Input | $1.35 / 1M tokens |
| Output | $2.70 / 1M tokens |
| Cached input | $1.35 / 1M tokens |
| Prompts of 200K tokens or more | |
| Input | $2.70 / 1M tokens |
| Output | $5.40 / 1M tokens |
| Cached input | $0.432 / 1M tokens |
Tiered pricing: when a request's prompt (input, cached tokens included) is 200K tokens or more, every token in that request is billed at the long-context rates. This model publishes no separate cached-input price for the base group: cached tokens there are billed at the base input rate (that group's "Cached input" row shows that rate).
| Release date (public API) | 2026-04-30docs.x.ai β | verified |
| API model slug | grok-4.3docs.x.ai β | verified |
| Key aliases (xAI side) | grok-4.3-latest, grok-latest, grok-4, grok-4-latest, grok-4-0709, grok-4-fast-reasoning, grok-4-1-fast-reasoning, grok-3, grok-3-mini (and more)docs.x.ai β | verified |
| Reasoning mode | Always on (chain-of-thought built in; no toggle required)codersera.com β | verified |
| Native video input | Not publicly documented via retrievable vendor URL | unknown |
| Native artifact generation | Not publicly documented via retrievable vendor URL | unknown |
| Training cutoff | Not publicly disclosed | unknown |
| Rate limits (xAI API) | 1,800 requests/min Β· 10,000,000 tokens/mindocs.x.ai β | verified |
| Regions | us-east-1, eu-west-1, us-west-2docs.x.ai β | verified |
| License | Proprietary | verified |
| Benchmark | Distribution | Score | Source |
|---|---|---|---|
Artificial Analysis Intelligence Index Artificial Analysis Intelligence Index v4.1.1 covering 9 evaluations (GDPval-AA v2, ΟΒ³-Banking, Terminal-Bench v2.1, SciCode, HLE, GPQA Diamond, CritPt, AA-Omniscience, AA-LCR). Measured and published by Artificial Analysis. Grok 4.3 sits below GPT-5.5 (60) and Gemini 3.1 Pro Preview (57) on this index, but at a fraction of their price. Costs $395 to run the full benchmark suite, ~20% lower than Grok 4.20. | 53points | artificialanalysis.ai β | |
GDPval-AA (agentic) Measured by Artificial Analysis on GDPval-AA v2 (agentic real-world task benchmark). Grok 4.3 scores 1500 Elo, up 321 points from Grok 4.20's 1179. Surpasses Gemini 3.1 Pro Preview, Muse Spark, GPT-5.4 mini (xhigh), and Kimi K2.5, but still trails GPT-5.5 (xhigh) by 276 Elo points. | 1500 Elo (+321 vs Grok 4.20 at 1179) | artificialanalysis.ai β | |
AA-Omniscience (factual accuracy) Measured by Artificial Analysis. Grok 4.3 gains 8 points on AA-Omniscience Accuracy versus Grok 4.20, but non-hallucination rate slipped 8 points in the same suite β Grok 4.20 still leads xAI's lineup on AA-Omniscience Non-Hallucination Rate. | +8 points accuracy vs Grok 4.20 | artificialanalysis.ai β | |
Long-horizon agentic tasks (community eval) A community evaluation by Towards AI (May 2026) claimed 13/18 tasks won versus Claude Opus 4.7 at ~9Γ lower cost. This is not a standardised benchmark and could not be independently verified from a retrievable official source. | β | Not publicly disclosed | β |
Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.
curl https://api.therouter.ai/v1/chat/completions -H "Content-Type: application/json" -H "Authorization: Bearer $THE_ROUTER_API_KEY" -d '{
"model": "xai/grok-4.3",
"messages": [
{"role": "user", "content": "Summarize the key points from this input."}
]
}'curl https://api.therouter.ai/v1/chat/completions \
-H "Authorization: Bearer $THEROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "xai/grok-4.3",
"messages": [{"role": "user", "content": "Explain the key differences between Rust and Go for a backend service."}]
}'xAI quietly released Grok 4.3 to the public API on April 30, 2026 with no press release or model card. The release was identified through API logs and quiet documentation updates. Key additions: 1M token context, native video input, always-on reasoning, native PDF/PowerPoint/spreadsheet generation, and a price cut of 37.5β83% versus Grok 4.20.
xAI began rolling out Grok 4.3 in beta to grok.com users and SuperGrok mobile subscribers on April 17, 2026, two weeks before the public API launch.
Yes β chain-of-thought reasoning is always on in Grok 4.3 and cannot be disabled through the API. This is a design change from Grok 4.20, where reasoning was a configurable parameter. The benefit is consistent multi-step quality without configuration; the trade-off is higher time-to-first-token (~11.6 seconds measured) compared to non-reasoning models. If your workload is latency-critical and does not need deep reasoning, route to xai/grok-3-mini or a fast non-reasoning xAI variant instead.
As of May 2026, xAI has consolidated a very large number of aliases β including grok-3, grok-3-mini, grok-4, grok-4-latest, grok-4-0709, grok-4-fast-reasoning, grok-4-1-fast-reasoning, and more β to all resolve to grok-4.3. This means if you call grok-4 directly on the xAI API, you get Grok 4.3, not the original July 2025 Grok 4. TheRouter preserves the original model behaviour: xai/grok-4 on TheRouter is pinned to the grok-4-0709 snapshot. Use xai/grok-4.3 explicitly on TheRouter if you want the current flagship.
Grok 4.3 (April 2026) improves on Grok 4.20 (March 2026) on several fronts: context doubled to 1M tokens, native video input added, +8 points factual accuracy (AA-Omniscience), +300 Elo on agentic tool-use (GDPval-AA), and prices cut 37.5β83%. The trade-off: non-hallucination rate dipped 8 points according to Artificial Analysis; Grok 4.20 still leads on that specific metric. Always-on reasoning also means higher TTFT in 4.3 vs Grok 4.20's configurable reasoning modes.
Grok 4.3 supports native video input (mp4/mov/webm, up to 5 minutes, 1080p) as a new modality. Whether TheRouter's proxy path forwards video input to the xAI API is subject to TheRouter's multimodal support status. Check TheRouter's model page or API documentation for the current video support status. For confirmed video support, calling the xAI API directly is the most reliable path.
| source | URL | retrieved | |
|---|---|---|---|
| Release date (public API) | docs.x.ai β | 2026-06-03 | verified |
| API model slug | docs.x.ai β | 2026-06-03 | verified |
| Key aliases (xAI side) | docs.x.ai β | 2026-06-03 | verified |
| Reasoning mode | codersera.com β | 2026-06-03 | verified |
| Native video input | β | β | unknown |
| Native artifact generation | β | β | unknown |
| Training cutoff | β | β | unknown |
| Rate limits (xAI API) | docs.x.ai β | 2026-06-03 | verified |
| Regions | docs.x.ai β | 2026-06-03 | verified |
| License | β | β | verified |
| Artificial Analysis Intelligence Index | artificialanalysis.ai β | 2026-08-19 | verified |
| GDPval-AA (agentic) | artificialanalysis.ai β | 2026-08-19 | verified |
| AA-Omniscience (factual accuracy) | artificialanalysis.ai β | 2026-08-19 | verified |
| Long-horizon agentic tasks (community eval) | pub.towardsai.net β | 2026-06-03 | unknown |
| Grok 4.3 launches on the public API | towardsai.net β | 2026-06-03 | verified |
| Grok 4.3 beta opens on grok.com and SuperGrok apps | codersera.com β | 2026-06-03 | verified |
| Is reasoning always on in Grok 4.3? Can I disable it? | codersera.com β | 2026-06-03 | to verify |
| What happened to grok-3, grok-4, and the other legacy aliases? | docs.x.ai β | 2026-06-03 | to verify |
| How does Grok 4.3 compare to Grok 4.20? | codersera.com β | 2026-06-03 | to verify |