Gemma 4 31B
Gemma 4 31B is served by 19 companies at prices ranging from $0.10 to $0.75 per million input tokens, a 7.5× difference for identical weights.
- cheapest host · CoreWeave
- $0.10
- price spread across hosts
- 7.5×
- companies serving it
- 19
- intelligence index
- 4.4
Is it worth it?
Beaten on both axes. GLM 5.3 Flash scores 27.8 against 4.4 and costs 1.3× less for the same work.
It beats GLM 5.2, DeepSeek V4.1 Flash at the same quality band, for more money.
Every host, compared
The same weights, priced and configured differently by 19 companies. Precision matters: 4-bit builds measurably degrade output and cost the same.
| Host | $/1M in | $/1M out | Cached | Context | Precision | Uptime |
|---|---|---|---|---|---|---|
| CoreWeavecheapest | $0.100 | $0.340 | $0.100 | 262K | fp4 | 100% |
| Venice | $0.120 | $0.360 | $0.090 | 256K | fp4 | 99.8% |
| Chutes | $0.120 | $0.370 | $0.012 | 131K | fp4 | 99.9% |
| DeepInfra | $0.135 | $0.510 | $0.075 | 262K | fp4 | 99.7% |
| DekaLLM | $0.140 | $0.400 | $0.050 | 262K | unknown | 95.7% |
| Crusoe | $0.140 | $0.400 | $0.140 | 262K | bf16 | 100% |
| Friendli | $0.140 | $0.400 | n/a | 262K | unknown | 100% |
| Novita | $0.140 | $0.400 | n/a | 262K | bf16 | 87.4% |
| NextBit | $0.150 | $0.400 | $0.060 | 262K | fp4 | 99.9% |
| Parasail | $0.150 | $0.400 | $0.060 | 262K | fp8 | 99.9% |
| DeepInfra | $0.160 | $0.320 | n/a | 262K | fp8 | 100% |
| DeepInfra | $0.200 | $0.400 | n/a | 262K | fp8 | 100% |
| DeepInfra | $0.300 | $0.600 | n/a | 262K | fp8 | 99.2% |
| Modal | $0.340 | $0.100 | $0.100 | 131K | unknown | 99.4% |
| Io Net | $0.361 | $1.09 | $0.181 | 262K | unknown | 93.1% |
| SambaNova | $0.380 | $1.15 | n/a | 131K | unknown | 98% |
| PrimeIntellect | $0.400 | $0.670 | $0.200 | 262K | fp4 | 99.3% |
| ModelRun | $0.750 | $1.00 | $0.200 | 262K | fp4 | 100% |
| SiliconFlow | $0.750 | $1.00 | $0.250 | 262K | fp8 | 96.5% |
7 of 19 hosts serve this model at 4-bit precision (CoreWeave, Venice, Chutes, DeepInfra, NextBit, PrimeIntellect, ModelRun), at the same price as full-precision hosts. See why this matters.
What it costs to run
- Marketing emails
- $0.200
- per 1,000 · short brief in, one email out
- Support replies
- $0.300
- per 1,000 · ticket plus help-centre context
- Lead enrichment
- $0.360
- per 1,000 · company page in, fields out
- Long-form drafts
- $0.840
- per 1,000 · outline in, full draft out
- Questions on docs
- $3.16
- per 1,000 · one large document, many questions
- Agent / tool loop
- $2.57
- per 1,000 · long context reused each step
Priced on the median host. Change the workload on the leaderboard to see how the ranking moves.
Specification
- Context, most hosts
- 262K · 393 pages
- Context range across hosts
- 131K to 262K
- Max output
- 16K
- Cache discount
- not published
- Input types
- image, text, video
- Tool use
- supported
- Reasoning
- optional
- Arena Elo
- n/a
- Agentic Elo
- n/a
- AA Intelligence
- 4.4
Other models with the same host spread
Host pricing and precision read live from OpenRouter’s endpoint API, compiled 9 November 2026. Free to quote with attribution. See all findings.