Grok 4.20 0309 (Reasoning)

Grok 4.20 0309 (Reasoning) combines a strong Intelligence Index position with unusually limited public product evidence. The available data places the model at rank 87 of 578, with an Intelligence Index score of 36.5. That position makes the model relevant for evaluation, but it does not establish that the model is dependable for a production workload.

The commercial signal is easier to read than the product story. The listed blended price is $3 per 1M tokens, with input priced at $2 and output priced at $6. Reported time to first token is 0.3 seconds. Median output speed is not reported, and the context window is also not reported.

The research brief found no verifiable official announcement, developer documentation, pricing page, benchmark material, or reliable community discussion for this model. That absence matters. Developers cannot confidently confirm whether Grok 4.20 0309 (Reasoning) remains directly callable, whether its alias is stable, or whether a newer model has replaced it.

The performance and pricing figures come from Artificial Analysis. Those figures support a screening decision. They do not answer operational questions about uptime, API behavior, tool use, structured output, rate limits, or regression risk.

Grok 4.20 0309 (Reasoning) offers a better Intelligence Index position than several nearby models while avoiding the highest listed blended prices in the comparison set. Its score of 36.5 is slightly above Claude 4.5 Sonnet (Reasoning) at 36.4, MiMo-V2-Omni-0327 at 36.4, and GPT-5 Codex (high) at 36.1. GPT-5.1 (high) is listed at 36.9.

That narrow spread changes the selection question. Grok 4.20 0309 (Reasoning) should not be treated as an automatic quality winner. The ranking says the model belongs in a serious shortlist. It does not show a large general-intelligence gap over its closest references.

Choice	Practical reading
Grok 4.20 0309 (Reasoning)	Lower blended cost than Claude 4.5 Sonnet (Reasoning) and MiMo-V2-Omni-0327, with a stronger Intelligence Index score than both listed references
GPT-5.1 (high)	Slightly higher Intelligence Index score, but a higher blended price and no reported median output speed
Gemini 3.5 Flash-Lite	Same Intelligence Index score, much lower listed blended price, and reported output speed of 381.175 tokens per second

The table is a decision aid, not a substitute for task testing. The research brief provides no verified explanation of Grok 4.20 0309 (Reasoning)'s intended role, preferred workloads, or known limitations. Developers should therefore interpret the model as a benchmark-led candidate rather than a fully documented platform choice.

Grok 4.20 0309 (Reasoning) has enough aggregate performance evidence to justify testing, but not enough task-specific evidence to justify default adoption. Rank 87 of 578 on the Artificial Analysis Intelligence Index indicates a comparatively strong position within the measured field. The result is useful for narrowing options before hands-on evaluation.

The ranking does not reveal how the model behaves on repository-scale coding, debugging, mathematics, extraction, planning, or tool-mediated workflows. The data brief does not provide a coding score, math score, context-window value, or output-throughput value for Grok 4.20 0309 (Reasoning). Those missing fields limit any claim about developer productivity.

The adjacent models show why aggregate rank needs context. Claude 4.5 Sonnet (Reasoning) has a coding index of 52.1 and a math index of 88 in the supplied comparison data. GPT-5.1 (high) has a coding index of 49.4 and a math index of 94. Gemini 3.5 Flash-Lite has a coding index of 49.3 and reported output speed of 381.175 tokens per second. Grok has no corresponding task-specific scores in the brief.

Grok 4.20 0309 (Reasoning) also reports 0.3 seconds to first token, matching the listed latency for each closest model. That makes initial responsiveness look competitive, but missing output speed prevents a complete interactive-performance judgment.

The correct performance conclusion is conditional: Grok deserves a place in a developer benchmark, especially for reasoning-heavy tasks, but the supplied evidence cannot confirm that it will outperform the nearby alternatives on code generation, code repair, or long-running agent workflows.

Grok 4.20 0309 (Reasoning) is priced as a moderate-cost reasoning model, with a listed blended rate of $3 per 1M tokens. The price is lower than Claude 4.5 Sonnet (Reasoning) at $6 and MiMo-V2-Omni-0327 at $15. It is also lower than GPT-5.1 (high) and GPT-5 Codex (high), each listed at $3.4375.

The input and output rates create an important workload distinction. Grok 4.20 0309 (Reasoning) lists $2 per 1M input tokens and $6 per 1M output tokens. Developers with large prompts and short answers may experience a different cost profile from developers running verbose reasoning, code generation, or agent loops. The blended figure is therefore a useful planning reference, not a universal invoice estimate.

Gemini 3.5 Flash-Lite is the strongest price reference in the supplied set. Its blended price is listed as $0.8500000000000001, with input at $0.3 and output at $2.5. Gemini also has the same Intelligence Index score of 36.5 and reported output speed of 381.175 tokens per second. Those facts make Grok less compelling for high-volume, latency-sensitive, or cost-first workloads unless Grok produces materially better task results in testing.

Grok 4.20 0309 (Reasoning) may still be cost-effective when stronger reasoning quality reduces retries, human review, or downstream repair. The supplied data does not measure those operational effects. No verified documentation confirms its availability, limits, or billing behavior beyond the listed data snapshot, so procurement should treat the price as a value to validate before committing volume.

Grok 4.20 0309 (Reasoning) is a sensible pilot candidate for teams that value aggregate reasoning performance at a $3 blended-token price. Rank 87 of 578 and an Intelligence Index score of 36.5 provide a credible reason to test it. The 0.3-second first-token latency also supports interactive evaluation.

The model is a better fit when the team can run its own acceptance suite and has a fallback provider. A useful pilot should test representative code changes, bug fixes, structured responses, refusal behavior, long prompts, retry rates, and tool calls. The evaluation should record completed-task quality, review effort, latency, and total tokens. Those measurements will answer questions that the public evidence does not.

Choose Grok when	Prefer another shortlist model when
A reasoning model with a $3 blended price fits the budget and independent validation is available	Cost and throughput dominate, making Gemini 3.5 Flash-Lite’s listed $0.8500000000000001 blended price and 381.175 output tokens per second more important
Aggregate Intelligence Index standing is the first screening criterion	Coding-specific evidence matters, because the brief reports coding scores for Claude 4.5 Sonnet (Reasoning), GPT-5.1 (high), and Gemini 3.5 Flash-Lite, but not Grok
The team can tolerate uncertainty about availability and model lifecycle	Stable documentation, confirmed access, and known failure modes are required before deployment

Do not choose Grok 4.20 0309 (Reasoning) solely because its aggregate score exceeds several adjacent models by a small margin. The research brief found no verifiable official source or reliable community evidence. Availability, alias stability, production limits, and failure behavior therefore remain open questions. The final recommendation is pilot first, commit only after access and task-level results are confirmed.

Grok 4.20 0309 (Reasoning) requires an evidence check before technical adoption because the available brief does not verify its current product status. The following questions focus on decisions that the aggregate score and price cannot settle.

What is the strongest evidence for considering Grok 4.20 0309 (Reasoning)?

The strongest evidence is its Artificial Analysis Intelligence Index position, where Grok 4.20 0309 (Reasoning) ranks 87 of 578 with a score of 36.5. That result supports shortlist inclusion, not automatic deployment.

What important performance information is missing?

The brief does not provide Grok 4.20 0309 (Reasoning)'s coding score, math score, context-window value, or median output speed. Developers must measure those dimensions with representative workloads before making a production decision.

Is Grok 4.20 0309 (Reasoning) a low-cost model?

Grok 4.20 0309 (Reasoning) is cheaper than several nearby reasoning models at a listed $3 blended price per 1M tokens. Gemini 3.5 Flash-Lite remains a much cheaper reference at $0.8500000000000001 blended tokens.

Does the model appear responsive?

Grok 4.20 0309 (Reasoning) has a reported 0.3-second time to first token, which matches the listed latency of its closest comparison models. Its output speed is not reported, so streaming completion speed remains unconfirmed.

Can developers confirm that the model is still available?

Developers cannot confirm current availability from the supplied research brief because no verifiable official announcement, developer document, pricing page, or stable-alias evidence was found. Access should be tested directly before planning a dependency.

Who should avoid adopting it first?

Teams that require verified documentation, stable naming, known failure modes, or coding-specific benchmark evidence should avoid making Grok 4.20 0309 (Reasoning) their first production dependency. The current evidence does not establish those conditions.