Price from DeepInfra: the cheapest offer that is serving, at standard delivery, with at least 90% of the reference context and at a declared precision that is not 4-bit (fp8), at its standard rate. The maker’s own precision is not known.
Cheaper right now: $0.080/M at DeepInfra — delivery tier (flex) It does not pass the like-for-like test, so it never sets a rank.
Every price dimension · USD per million tokens
Input
$0.080/M
Output
$0.160/M
Cached input
Unknown/M
Cache write
Unknown/M
Cache write 1h
Unknown/M
Reasoning
Unknown/M
Who sells it
The same weights, different shops. Cheapest is not like-for-like when serving precision differs.
6 provider offers · rates, limits and conditions
Rates are USD per million tokens. Endpoint terms can differ even when the quoted price and precision match. A listed parameter is a provider declaration, not a task-success test.
Seller
Input
Output
Precision
Uptime (1d)
Seller
DeepInfradeepinfra/fp8/flex Endpoint terms
Cached input /M
Unknown
Context limit
131,072 tokens
Output limit
117,964 tokens
Tools
Listed by endpoint
Reasoning
Not listed by endpoint
Promotional discount
None reported
Uptime · last 30 minutes
Unknown
Input$0.064
Output$0.128
Precisionfp8
Uptime (1d)100.00%
Seller
DeepInfradeepinfra/fp8 Endpoint terms
Cached input /M
Unknown
Context limit
131,072 tokens
Output limit
16,384 tokens
Tools
Listed by endpoint
Reasoning
Not listed by endpoint
Promotional discount
None reported
Uptime · last 30 minutes
100.00%
Input$0.080
Output$0.160
Precisionfp8
Uptime (1d)99.93%
Seller
NextBitnextbit/fp8 Endpoint terms
Cached input /M
$0.040
Context limit
131,072 tokens
Output limit
117,964 tokens
Tools
Not listed by endpoint
Reasoning
Not listed by endpoint
Promotional discount
None reported
Uptime · last 30 minutes
96.67%
Input$0.080
Output$0.350
Precisionfp8
Uptime (1d)98.11%
Seller
Parasailparasail/fp8 Endpoint terms
Cached input /M
$0.040
Context limit
131,072 tokens
Output limit
117,964 tokens
Tools
Not listed by endpoint
Reasoning
Not listed by endpoint
Promotional discount
None reported
Uptime · last 30 minutes
100.00%
Input$0.080
Output$0.450
Precisionfp8
Uptime (1d)100.00%
Seller
Nebiusnebius/fp8 Endpoint terms
Cached input /M
$0.040
Context limit
110,000 tokens
Output limit
99,000 tokens
Tools
Not listed by endpoint
Reasoning
Not listed by endpoint
Promotional discount
None reported
Uptime · last 30 minutes
100.00%
Input$0.100
Output$0.300
Precisionfp8
Uptime (1d)99.19%
Seller
DeepInfradeepinfra/fp8/priority Endpoint terms
Cached input /M
Unknown
Context limit
131,072 tokens
Output limit
117,964 tokens
Tools
Listed by endpoint
Reasoning
Not listed by endpoint
Promotional discount
None reported
Uptime · last 30 minutes
100.00%
Input$0.120
Output$0.240
Precisionfp8
Uptime (1d)99.92%
Independent scores
LMArena Elo
1357.8default
LMArena row
gemma-3-27b-it
LMArena Elo under CC BY 4.0. Absence of another board is not a score of zero.
Gemma 3 introduces multimodality, supporting vision-language input and text outputs. It handles context windows up to 128k tokens, understands over 140 languages, and offers improved math, reasoning, and chat capabilities,...
Questions this page answers
What does Gemma 3 27B cost?
$0.080/M in, $0.160/M out in this catalogue, as of the fetch date on this page. That is the API row, not a subscription.
Does Gemma 3 27B have an independent quality score?
Gemma 3 27B has an LMArena score in this catalogue.
What beats Gemma 3 27B?
gpt-oss-120b has an equal-or-higher measured score and an equal-or-lower price, with capability trade-offs.
Does this page use Artificial Analysis scores for Gemma 3 27B?
No. Artificial Analysis figures are not published here. Quality on this page is LMArena Elo where a score exists; otherwise the row is unrated.
MODEL MONUMENT
Save or share this proof
Preview and more sharing options
Google: Gemma 3 27B. Balanced workload. Alternative has a trade-off. Evidence as of 22 Nov 2026.
Related
Higher score or lower price, with a named tradegpt-oss-120b
The verdict as one SVG, for a README or a docs page. It states this model's dominance
status at the balanced workload on the LMArena lens, with the date it was computed, and
it is rebuilt with the catalogue — so it changes when the verdict changes, including to
one you would rather it did not.