Grok 4.5 leads VulcanBench’s new coding benchmark with 91.3%, solving…

What was claimed

Grok 4.5 leads VulcanBench’s new coding benchmark with 91.3%, solving 21 of 23 real-world software tasks, beating Claude Fable 5 and GPT-5.6 Sol while owning the cost-efficiency frontier

Our verdict

Inaccurate

Across neutral coding benchmarks, Claude Fable 5 consistently scores higher than Grok 4.5: for example, DeepSWE 1.1 (70% vs 53%), SWE Bench Pro (80.4% vs around mid‑field for Grok 4.5), and aggregate coding leaderboards where Fable 5 tops charts and Grok trails frontier models. These sources show Grok 4.5 does not generally beat Fable 5 in coding accuracy. Sources agree Grok 4.5 is highly cost‑efficient and more token‑efficient than many frontier models, with pricing at $2/M input and $6/M output and far fewer output tokens per SWE Bench Pro task than Opus 4.8. However, other models like GPT‑5.6 Luna and Terra have competitive or lower input/output prices, and analyses explicitly state that Grok 4.5 trails frontier models in accuracy, so claiming it "owns" the entire cost‑efficiency frontier over all competitors overstates and simplifies the situation.

2 of 3 AI systems agree13 sources citedChecked Jul 20, 2026

Check your own claim

Paste any statement, headline, or AI answer — 3 independent AIs verify it in seconds, with sources.

Key findings

Grok 4.5 beat Claude Fable 5 and GPT-5.6 Sol on that VulcanBench report.

Incorrect88%

2 of 3 AIs agree·ChatGPT: Verified

Grok 4.5 "owns the cost-efficiency frontier."

Misleading70%

All 2 AIs agree

Grok 4.5 leads VulcanBench’s new coding benchmark with 91.3%, solving 21 of 23 real-world software tasks.

Can’t verify60%

2 of 3 AIs agree·ChatGPT: Verified

Verified against 13 sources

Detailed Analysis

The response makes several concrete benchmark and ranking claims that conflict with current published data. Key statements about Grok 4.5 leading VulcanBench, scoring 91.3%, beating Claude Fable 5 and GPT-5.6 Sol, and owning the cost-efficiency frontier cannot be verified and are often contradicted by available sources. Overall, the answer is factually unreliable and misleading.

Why this verdict

The response makes several concrete benchmark and ranking claims that conflict with current published data.
Key statements about Grok 4.5 leading VulcanBench, scoring 91.3%, beating Claude Fable 5 and GPT-5.6 Sol, and owning the cost-efficiency frontier cannot be verified and are often contradicted by available sources.
Overall, the answer is factually unreliable and misleading.

Claims checked

Grok 4.5 beats GPT-5.6 Sol

Incorrect85%

On multi‑domain and coding leaderboards cited in the search results, Grok 4.5 ranks below several OpenAI models including GPT‑5.6 Sol, and its coding score is the lowest of any model listed.[8] [13] This contradicts the claim that Grok 4.5 beats GPT‑5.6 Sol overall in coding benchmarks.

This claim may mislead readers without added context.

Grok 4.5 beats Claude Fable 5

Incorrect90%

This claim may mislead readers without added context.

Grok 4.5 owns the cost-efficiency frontier

Misleading80%

Sources agree Grok 4.5 is highly cost‑efficient and more token‑efficient than many frontier models, with pricing at $2/M input and $6/M output and far fewer output tokens per SWE Bench Pro task than Opus 4.8.[3] [4] [9] [10] [12] [13] However, other models like GPT‑5.6 Luna and Terra have competitive or lower input/output prices, and analyses explicitly state that Grok 4.5 trails frontier models in accuracy, so claiming it "owns" the entire cost‑efficiency frontier over all competitors overstates and simplifies the situation.[5] [11] [13]

This claim may mislead readers without added context.

What other AI think

The claim contains multiple significant factual errors. "VulcanBench" does not exist as a real benchmark. The 91.3% figure refers to tokens-per-second speed, not a benchmark score. The "21 of 23 tasks" metric is unverifiable. Most critically, Grok 4.5 actually trails Fable 5 and GPT-5.5 on real coding benchmarks, contradicting the claim that it "leads" them.

Why this verdict

The claim contains multiple significant factual errors.
"VulcanBench" does not exist as a real benchmark.
The 91.3% figure refers to tokens-per-second speed, not a benchmark score.
The "21 of 23 tasks" metric is unverifiable.

The numeric accuracy about VulcanBench’s report is verifiable and matches published results. However the phrase about "owning the cost-efficiency frontier" is subjective and not strictly a factual claim without more specific metrics or thresholds.

Why this verdict

The numeric accuracy about VulcanBench’s report is verifiable and matches published results.
However the phrase about "owning the cost-efficiency frontier" is subjective and not strictly a factual claim without more specific metrics or thresholds.

Claims checked

Grok 4.5 "owns the cost-efficiency frontier."

Misleading60%

VulcanBench reports cost and token metrics but "owning the cost-efficiency frontier" is an interpretive claim requiring a defined metric or Pareto analysis; the site notes Grok is efficient but doesn’t unambiguously prove exclusive dominance.

This claim may mislead readers without added context.

Grok 4.5 beat Claude Fable 5 and GPT-5.6 Sol on that VulcanBench report.

Verified88%

The same VulcanBench report compares Grok 4.5 against Fable 5 and GPT‑5.6 Sol and shows Grok leading on accuracy for that v1-micro suite.

Grok 4.5 leads VulcanBench’s new coding benchmark with 91.3%, solving 21 of 23 real-world software tasks.

Verified92%

VulcanBench’s Report07 (2026-07-12) lists a Grok 4.5 result of about 91% on a 23-task suite, which corresponds to 21/23 (≈91.30%).

Share this result

Try DoubleChecks