Is Granite 4.2 8B better than Phi 4?

On accuracy they are statistically tied: Granite 4.2 8B scored 36.2% and Phi 4 41.2% on the same 80 questions from open datasets. Similar accuracy; Granite 4.2 8B costs less per question.

Which is cheaper, Granite 4.2 8B or Phi 4?

Granite 4.2 8B is cheaper: $0.000077 per question as billed, against $0.00012. List prices: Granite 4.2 8B $0.06 in / $0.25 out, Phi 4 $0.07 in / $0.14 out per million tokens.

Which is faster, Granite 4.2 8B or Phi 4?

Phi 4 is faster at 103 output tokens per second, against 60. Time to first token: 0.6 s for Granite 4.2 8B, 0.5 s for Phi 4.

Which has the larger context window, Granite 4.2 8B or Phi 4?

Granite 4.2 8B, with 131k tokens against 16k.

Head to head · tested 27 Oct 2026

Granite 4.2 8B vs Phi 4: accuracy, speed and price

Similar accuracy; Granite 4.2 8B costs less per question.

IBMGranite 4.2 8B36.2% score · #60$0.000077 per question60 tok/s

MicrosoftPhi 441.2% score · #54$0.00012 per question103 tok/s

Verdict

Built from the measurements below

Accuracy is too close to call: Granite 4.2 8B scored 36.2% and Phi 4 41.2% on the same 80 questions, a gap within what chance alone could produce.
Granite 4.2 8B is 1.6× cheaper per question as billed ($0.000077 vs $0.00012), reasoning tokens included.
Phi 4 generates text 1.7× faster: 103 tokens per second against 60.
Phi 4 starts answering sooner: 0.5 s to the first token, against 0.6 s for Granite 4.2 8B.
Granite 4.2 8B accepts more context: 131k tokens against 16k.

Which should you choose?

Choose Granite 4.2 8B for the lower cost per correct answer ($0.00021 against $0.00029), shorter outputs (about 227 tokens per question against 699) and a 131k-token context window. Choose Phi 4 for faster streaming and a quicker first token. The widest subject gap is in medicine: Phi 4 answered 4 of 11 correctly, Granite 4.2 8B 1 of 11. With so few questions per subject, treat this as a lead to test on your own prompts, not a result.

Accuracy gap -5.0 points (z = -0.65). A winner is called only when z is at least 1.96, the usual 95% threshold. Speed, latency and price need a 10% margin.

Head to head

Same questions, same prompt, same API

Measure	Granite 4.2 8B	Phi 4	Edge
Verification scoreHigher is better	36.2%	41.2%	Too close to call
Cost per questionLower is better	$0.000077	$0.00012	Granite 4.2 8B
Output speedHigher is better	60	103	Phi 4
Time to first tokenLower is better	0.6 s	0.5 s	Phi 4
End-to-end response timeLower is better	7.3 s	4.9 s	Phi 4
Tokens per questionLower is better	227	699	Granite 4.2 8B
Context windowHigher is better	131k	16k	Granite 4.2 8B

Side by side

Bars are scaled to the larger of the two values. The winner on each measure is marked in bold.

Every measure

Granite 4.2 8B first, Phi 4 second

Verification scorehigher is better

Granite 4.2 8B36.2%

Phi 441.2%

Cost per questionlower is better

Granite 4.2 8B$0.000077

Phi 4$0.00012

Output speedhigher is better

Granite 4.2 8B60

Phi 4103

Time to first tokenlower is better

Granite 4.2 8B0.6 s

Phi 40.5 s

End-to-end response timelower is better

Granite 4.2 8B7.3 s

Phi 44.9 s

Tokens per questionlower is better

Granite 4.2 8B227

Phi 4699

Context windowhigher is better

Granite 4.2 8B131k

Phi 416k

Score vs. cost per question

Both models in color, every other model in grey

Results by subject and dataset

Questions answered correctly. Small samples: a gap of one or two questions is noise.

Subject	Questions	Granite 4.2 8B	Phi 4
Science	30	13/30	12/30
Engineering	15	4/15	6/15
Medicine	11	1/11	4/11
Economics	9	5/9	5/9
Law	6	2/6	2/6
Humanities & social science	9	4/9	4/9
SuperGPQA dataset	50	11/50	12/50
MMLU-Pro dataset	30	18/30	21/30

Specifications and pricing

From OpenRouter's catalogue, plus the leaderboard's test dates

Specification	Granite 4.2 8B	Phi 4
Creator	IBM	Microsoft
Released	31 Aug 2026	10 Jan 2025
Weights	Open weights	Open weights
Reasoning	Yes	No
Input	text	text
Output	text	text
Context window	131k tokens	16k tokens
Knowledge cutoff	—	30 Jun 2024
List price in / out per 1M	$0.06 / $0.25	$0.07 / $0.14
Blended price per 1M (3:1)	$0.11	$0.09
Cheapest host	CoreWeave · $0.06 / $0.15	DeepInfra · $0.056 / $0.112
Hosts on OpenRouter	2	1
Last tested	26 Oct 2026	26 Oct 2026

Compare any two models →

Granite 4.2 8B vs Phi 4: frequently asked questions

Is Granite 4.2 8B better than Phi 4?: On accuracy they are statistically tied: Granite 4.2 8B scored 36.2% and Phi 4 41.2% on the same 80 questions from open datasets. Similar accuracy; Granite 4.2 8B costs less per question.
Which is cheaper, Granite 4.2 8B or Phi 4?: Granite 4.2 8B is cheaper: $0.000077 per question as billed, against $0.00012. List prices: Granite 4.2 8B $0.06 in / $0.25 out, Phi 4 $0.07 in / $0.14 out per million tokens.
Which is faster, Granite 4.2 8B or Phi 4?: Phi 4 is faster at 103 output tokens per second, against 60. Time to first token: 0.6 s for Granite 4.2 8B, 0.5 s for Phi 4.
Which has the larger context window, Granite 4.2 8B or Phi 4?: Granite 4.2 8B, with 131k tokens against 16k.