Direct answer: at Azure Managed Redis public list prices on 10 November 2026 (eastus, USD, excluding tax and discounts), Balanced instances bill hourly: B0 $0.016, B10 $0.315, B20 $0.629, B50 $1.257, B100 $2.496, B250 $5.04, B500 $9.984, B1000 $19.968; HA (primary + replica, two nodes) is 2× the single-node price. B10 (12 GB/4 vCPU) is $229.95/month single-node and $459.90 HA; 1-year reserved HA is $149.50/month ($1,794/year), 3-year $103.47/month ($3,725/3 years). Per GB HA: Balanced about $30.4, Memory Optimized about $21, Compute Optimized about $56.7, Flash Optimized about $8.7. Single-node is dev/test only; production should use two-node HA, and reservations cut B10 from $459.90 to $149.50/$103.47. Flash is cheapest but does not support RediSearch/Bloom/TimeSeries; B0/B1 lack active geo-replication; >350 GB in-memory tiers (B500+/X500+/M500+) and A2000/A4500 remain in preview.
This guide is for architects, app/SRE, finance and procurement choosing Azure cache, migrating from Azure Cache for Redis, or sizing across Balanced/Compute/Memory/Flash. It separates official prices, editorial calculations and exclusions, covering the four tiers, HA 2×, single vs two nodes, $/GB across tiers, 1/3-year reservations, hidden costs and a 90-day checklist. Prices come from the Managed Redis pricing page, Azure Retail Prices API and Microsoft Learn (10 November 2026). This is not procurement advice.
(1) HA is 2× the node price, not a free replica: budget two nodes for production; single-node is dev/test only. (2) Reservations use the HA total: B10 drops from $459.90/month to $149.50 (1-year) or $103.47 (3-year), but requires commitment. (3) Choose by $/GB and modules: Flash is cheapest at ~$8.7/GB but lacks RediSearch/Bloom/TimeSeries; Compute is most expensive at ~$56.7/GB and buys throughput.
Four tiers: ratio, capacity and modules
Azure Managed Redis has four tiers: Memory Optimized 8:1 (12–1,920 GB, large memory/dev-test), Balanced 4:1 (0.5–960 GB, standard workloads), Compute Optimized 2:1 (3–720 GB, maximum throughput), Flash Optimized (250–4,500 GB, NVMe + RAM, low-cost large capacity). All support Private Link, Entra ID, replication/failover, zone redundancy, persistence, clustering and import/export; Flash does not support RediSearch/Bloom/TimeSeries (JSON is supported), and B0/B1 lack active geo-replication.
| Tier | Memory:vCPU | Capacity | Fit | Note |
|---|---|---|---|---|
| Memory Optimized | 8:1 | 12–1,920 GB | Large memory, dev/test | M500+ preview |
| Balanced | 4:1 | 0.5–960 GB | Standard workloads | B500+ preview; B0/B1 no active geo |
| Compute Optimized | 2:1 | 3–720 GB | Maximum throughput | X500+ preview |
| Flash Optimized | NVMe+RAM | 250–4,500 GB | Large capacity, low cost | No Search/Bloom/TimeSeries; A2000/A4500 preview |
Sources: Managed Redis pricing page and Learn "Azure Managed Redis overview" (10 November 2026). HA deploys primary + replica across at least two nodes; >350 GB in-memory tiers are preview; features follow the official docs.
Balanced: single node, HA and reservations
Balanced is the common 4:1 tier. B0 (1 GB/2 vCPU) is $11.68/month single-node, $23.36 HA; B10 (12/4) $229.95/$459.90; B20 (24/8) $459.17/$918.34; B50 (60/16) $917.61/$1,835.22; B100 (120/32) $1,822.08/$3,644.16; B250 (235/64) $3,679.20/$7,358.40; B500 (480/128) $7,288.32/$14,576.64; B1000 (960/256) $14,576.64/$29,153.28. 1-year reserved HA is about 32%–33% of HA PAYG; 3-year about 22%–23%; B10 is $149.50/month (1-year) and $103.47 (3-year).
| SKU | GB/vCPU | Single/month | HA/month | 1-year HA/month | 3-year HA/month |
|---|---|---|---|---|---|
| B0 | 1 / 2 | $11.68 | $23.36 | $7.58 | $5.25 |
| B10 | 12 / 4 | $229.95 | $459.90 | $149.50 | $103.47 |
| B20 | 24 / 8 | $459.17 | $918.34 | $298.50 | $206.64 |
| B50 | 60 / 16 | $917.61 | $1,835.22 | $596.42 | $412.92 |
| B100 | 120 / 32 | $1,822.08 | $3,644.16 | $1,184.33 | $819.94 |
| B250 | 235 / 64 | $3,679.20 | $7,358.40 | $2,391.50 | $1,655.64 |
| B500 | 480 / 128 | $7,288.32 | $14,576.64 | $4,737.42 | $3,279.75 |
| B1000 | 960 / 256 | $14,576.64 | $29,153.28 | $9,474.83 | $6,559.50 |
Editorial calculation: PAYG hourly × 730; HA = 2× single; 1/3-year = official reservation total ÷ 12/36 (HA); excludes tax, discounts, Private Link/egress. B500+ preview.
Compute vs Balanced: about $56.7 vs $30.4 per GB
Compute Optimized is 2:1, with more CPU and a higher $/GB: X3 (3 GB/4) HA $319.74/month, $106.58/GB; X10 (12/8) $851.18, $70.93/GB; X50 (60/32) $3,403.26, $56.72/GB; X100 (120/64) $6,807.98, $56.73/GB. Balanced HA is about $30.4/GB, roughly 54% of Compute. Only move to Compute when throughput/connections really require it; standard caching is cheaper on Balanced.
| SKU | GB/vCPU | HA/month | $/GB | Balanced comparison |
|---|---|---|---|---|
| X3 | 3 / 4 | $319.74 | $106.58 | B3 from $23.36/GB |
| X10 | 12 / 8 | $851.18 | $70.93 | B10 $38.33 |
| X50 | 60 / 32 | $3,403.26 | $56.72 | B50 $30.59 |
| X100 | 120 / 64 | $6,807.98 | $56.73 | B100 $30.37 |
Editorial calculation: PAYG hourly × 730 × 2 (HA); $/GB = HA ÷ GB; excludes reservations, tax and discounts. X500+ preview.
Memory vs Balanced: about $21 vs $30.4 per GB
Memory Optimized is 8:1, with more memory and less CPU, at a lower $/GB: M10 (12 GB/2) HA $315.36/month, $26.28/GB; M50 (60/8) $1,258.52, $20.98/GB; M100 (120/16) $2,517.04, $20.98/GB; M500 (480/64) $10,091.52, $21.02/GB. It suits large caches, sessions, leaderboards and read-heavy workloads; if the workload needs high throughput/connections, Balanced or Compute fits better. Some M SKUs lack public reservations and are priced PAYG.
| SKU | GB/vCPU | HA/month | $/GB | 1-year HA/month |
|---|---|---|---|---|
| M10 | 12 / 2 | $315.36 | $26.28 | — |
| M50 | 60 / 8 | $1,258.52 | $20.98 | — |
| M100 | 120 / 16 | $2,517.04 | $20.98 | — |
| M500 | 480 / 64 | $10,091.52 | $21.02 | $3,279.75 |
Editorial calculation: PAYG × 730 × 2 (HA); $/GB = HA ÷ GB; some M SKUs have no public reservation ("—"); M500+ preview.
Flash: about $8.7 per GB, but missing modules
Flash Optimized moves cold data to NVMe and keeps hot data in RAM, giving the lowest $/GB: A250 (256 GB/8) HA $2,277.60/month, $8.90/GB; A500 (525/16) $4,555.20, $8.68/GB; A1000 (1,050/32) $9,110.40, $8.68/GB; A4500 (4,723/144) $36,432.84, $7.71/GB. 1-year reserved A250 is $740.25/month and A4500 $11,840.67/month. The trade-off: no RediSearch/Bloom/TimeSeries and no non-clustered instances; if you need search, probabilistic structures or time series, Flash is not an option.
| SKU | GB/vCPU | HA/month | $/GB | 1-year HA/month |
|---|---|---|---|---|
| A250 | 256 / 8 | $2,277.60 | $8.90 | $740.25 |
| A500 | 525 / 16 | $4,555.20 | $8.68 | $1,480.42 |
| A1000 | 1,050 / 32 | $9,110.40 | $8.68 | $2,960.92 |
| A4500 | 4,723 / 144 | $36,432.84 | $7.71 | $11,840.67 |
Editorial calculation: PAYG × 730 × 2 (HA); 1-year = reservation total ÷ 12 (HA); $/GB = HA ÷ GB; A2000/A4500 preview; excludes tax and discounts.
Selection matrix and hidden costs
Choose by ratio/modules: standard caching uses Balanced; large-memory/read-heavy uses Memory; high throughput/many connections uses Compute; large capacity with cost priority and no Search/Bloom/TimeSeries uses Flash. Production uses two-node HA (2×); single-node is dev/test only. Seven hidden costs: HA 2×, reservation commitment, preview limitations, Flash module gaps, B0/B1 no active geo, Private Link/egress, and persistence/backup. 90 days: weeks 1–2 inventory data size, hit rate, connections/throughput and modules; weeks 2–4 price Balanced/Memory/Compute/Flash HA; weeks 4–6 compare 1/3-year reservations; weeks 6–8 load-test throughput/latency; weeks 8–10 verify zone redundancy/geo; weeks 10–12 reconcile nodes, reservations and egress against the bill.
| Scenario | First choice | Check first |
|---|---|---|
| Standard cache/session | Balanced | HA 2×, memory hit rate |
| Large memory/read-heavy | Memory | $/GB, enough CPU |
| High throughput/connections | Compute | Throughput/latency test |
| Large capacity/cost-first | Flash | No Search/Bloom/TimeSeries |
| Stable 1–3 years | HA + reservation | Commitment, SKU availability |
| Dev/test | Single node | No production SLA |
Editorial summary, not a vendor recommendation; features/regions/preview status follow the official docs.
FAQ
How is Azure Managed Redis billed?
Per cache-instance hour: Balanced B0 $0.016, B10 $0.315, B50 $1.257, B100 $2.496, B250 $5.04, B500 $9.984, B1000 $19.968; HA is 2×; optional 1/3-year reservations are HA totals.
Why is HA 2×?
HA deploys primary + replica on at least two nodes (across zones where supported), so it bills two instances; single-node is dev/test only. Production cache budgets should use HA.
What does B10 cost per month?
Single-node $229.95; HA $459.90; 1-year reserved HA $149.50/month, 3-year $103.47/month. Reservations save about 67%/77%, but require commitment.
Balanced, Memory or Compute?
Balanced is 4:1 for standard workloads; Memory is 8:1 for large memory/read-heavy at about $21/GB; Compute is 2:1 for high throughput at about $56.7/GB. Choose by memory/CPU ratio and throughput needs.
Is Flash cheaper?
$/GB is about $8.7, the lowest; but it only supports JSON, not RediSearch/Bloom/TimeSeries, and has no non-clustered instances. If you need search/probabilistic/time-series, Flash doesn't fit.
Are reservations worth it?
1-year is about 32%–33% of HA PAYG, 3-year about 22%–23%; worth it for stable 1–3-year workloads with a fixed SKU. Reservations use the HA total—confirm node count and regional availability first.
What are the B0/B1 limitations?
B0/B1 do not support active geo-replication and may lack some features; they suit small/dev-test use. Production should start at B10 or above.
How do I reduce cost?
Five steps: (1) use HA but the right tier—don't default to Compute; (2) size by hit rate/data volume, not overprovision; (3) buy 1/3-year reservations for stable load; (4) use Flash for large capacity when search isn't needed; (5) reconcile nodes, reservations and Private Link/egress.
Research method and limitations
Official facts (pricing page, Retail API, Learn): Balanced/Compute/Memory/Flash specs and hourly prices; HA = primary + replica two nodes; 1/3-year reservations are HA totals; Flash lacks Search/Bloom/TimeSeries; B0/B1 lack active geo; >350 GB in-memory and A2000/A4500 are preview. Editorial calculations: 730 hours/month, HA = 2×, $/GB = HA ÷ GB, reservation total ÷ 12/36.
Seven limitations: (1) only eastus public list prices, excluding tax/contract discounts; (2) capacity/hit/throughput are editorial assumptions; (3) some M SKUs have no public reservation; (4) excludes Private Link/egress, persistence/backup and tax; (5) no real throughput/latency/failover benchmark; (6) preview features/regions may change; (7) prices/SKUs change—recheck official pages and bills before purchase.
Sources and basis
- Microsoft Azure, Azure Managed Redis Pricing: azure.microsoft.com/en-us/pricing/details/managed-redis. Four tiers, specs, PAYG/HA/reservations, feature differences. Accessed 10 November 2026.
- Azure Retail Prices API (productName contains Azure Managed Redis, eastus): prices.azure.com/api/retail/prices. B/X/M/A hourly and 1/3-year reservation totals. Accessed 10 November 2026.
- Microsoft Learn, Azure Managed Redis overview: azure/redis/overview. Tier ratios, capacity, HA two nodes, Flash module limits, preview tiers. Accessed 10 November 2026.
Data cut-off 10 November 2026; prices/availability follow Microsoft official pages and bills; this is not procurement, commercial or legal advice.