Updated 4 Sep 2026: Adds material current pricing details from Ollama's live pricing page: DeepSeek V4 Flash/Pro rates double during weekday 12:00–18:00 UTC peak pricing, paid plans have explicit concurrency ceilings, and Team is currently labeled early access. This qualifies the original fixed-rate/limit framing.
Ollama replaces GPU-time subscriptions with transparent token pricing
Ollama’s cloud Pro, Max and Team plans now use token-priced usage credits instead of GPU-time billing, but the live pricing page adds two important constraints: selected DeepSeek models cost twice as much during weekday peak hours and plans have explicit concurrency ceilings.
Pro is $20/month with $60 of included monthly usage; Max is $100 with $300; Team starts at $500 with $1,000 shared across unlimited users.
Usage is metered by model-specific input, cached-input and output token rates.
DeepSeek V4 Flash and DeepSeek V4 Pro currently have peak pricing from 12:00–18:00 UTC Monday–Friday.
For those two models, the current peak input, cached-input and output rates are double their normal listed rates.
Current concurrency limits are Free 1, Pro 3, Max 10 and Team 10 concurrent requests.
Requests above concurrency are queued and can be rejected when the queue is full.
Existing subscribers retain legacy terms until they change billing cycle or tier.
The current pricing page labels Team 'Early access' while the launch post says it is available for signup.
What builders should take away
Instrument spend by model and time window; workloads using DeepSeek V4 Flash or Pro can cost twice as much during Ollama’s published weekday peak hours.
Track concurrency as well as dollar credits for coding-agent and multi-agent workloads; unused credits do not guarantee immediate parallel capacity.
Compare Ollama’s live token-rate schedule with direct provider APIs and self-hosting using the same prompt/output mix and traffic timing.
For Team, set shared balances and disable automatic usage billing where appropriate so one workload cannot silently consume the organization’s extra-credit balance.
If you are on legacy Pro or Max pricing, model the consequences before changing tier or billing cycle because that change moves the account to the new token-priced model.
What changed
Ollama announced August 31, 2026 that new Pro, Max and Team plans use per-token pricing with monthly usage-credit pools. Pro costs $20/month with $60 of monthly usage, Max costs $100 with $300, and Team starts at $500 with $1,000 shared across unlimited users. The current pricing page now makes the operating constraints more explicit than the launch post: DeepSeek V4 Flash and DeepSeek V4 Pro use peak pricing between 12:00 and 18:00 UTC Monday through Friday, with their listed input, cached-input and output rates doubling during that window. Ollama also documents concurrency limits of 1 request on Free, 3 on Pro, and 10 on Max and Team; excess requests queue until capacity is available and can be rejected if the queue is full. Existing subscribers can remain on legacy terms until they change tier or billing cycle.
Why it matters
The move away from GPU-time billing still gives builders a much clearer unit for comparing hosted inference, but 'published per-token pricing' now needs to be interpreted as a rate schedule rather than one fixed price for every model at every time. Agent workloads can also hit concurrency before they exhaust monthly credits. Teams comparing Ollama with direct provider APIs or self-hosting should therefore model token mix, time-of-day exposure and parallelism together instead of treating the monthly credit pool as the only constraint.
A move from hardware time to model usage
Ollama says the change responds to feedback that GPU-time billing was hard to predict, especially as open models became larger. The new plans expose model-by-model token rates and a monthly included-usage pool, making the main cost unit visible before a team commits to a workload.
Two DeepSeek models currently have weekday peak pricing
Ollama’s live pricing table lists separate peak rates for DeepSeek V4 Flash and DeepSeek V4 Pro between 12:00 and 18:00 UTC Monday through Friday. Each displayed input, cached-input and output rate is twice the normal listed rate during that window. Other models on the current table do not show a separate peak schedule.
Concurrency is a separate capacity limit
The current FAQ says Free supports one concurrent request, Pro three, and Max and Team ten. Requests beyond the plan limit are queued, and a request can be rejected if that queue reaches its own limit. This means a plan with unused dollar credits can still throttle a highly parallel agent workload.
Shared pools still change team economics
Team starts at $500/month with $1,000 of shared usage across unlimited users, centralized billing and administration. The live pricing page currently labels Team 'Early access', even though the August 31 launch post says Team is available for signup. That is a rollout-status wording inconsistency rather than a different price.
Migration remains opt-in for existing subscribers
Existing Pro, Max and Team subscribers can keep their legacy terms while they continue renewing. Changing billing cycle or tier moves the account to the new pricing model. Switching resets usage to the new plan’s full monthly amount and removes the old session and weekly limits.
Timeline
2026-08-31
Ollama launches token-priced plans
Pro, Max and Team move from GPU-time-style limits to monthly usage credits consumed at published token rates.
2026-09-04
Live pricing review exposes peak and concurrency constraints
Ollama’s current pricing page lists weekday peak rates for two DeepSeek models and explicit concurrency ceilings across plans.
What to watch next
Whether peak pricing expands to additional models or changes hours/rate multipliers.
Whether Ollama adds priority tiers, since the current pricing FAQ says faster priority tiers may be available in the future.
Whether concurrency limits or queue behavior change as Fast mode and Team features roll out.
Whether Team leaves its current 'Early access' label and gains the promised shared projects, skills and instructions.
Usage dashboards, API-level budget enforcement and regional routing/data-retention changes.
Still unclear
The pricing announcement is first-party and does not independently validate effective cost or throughput across workloads.
Peak pricing is currently shown only for DeepSeek V4 Flash and Pro; Ollama does not state on the pricing page whether other models could gain peak schedules later.
The current pricing page labels Team early access while the launch post says the plan is available for signup.
Actual savings versus legacy GPU-time billing depend on model mix, traffic timing, prompt/output length and concurrency.
K2 Horizon is notable less for another benchmark claim than for reproducibility: IFM is publishing model weights, architecture, training code, data or construction recipes, evaluation resources and intermediate training material instead of stopping at a final checkpoint.
Periskope is moving toward a hybrid SaaS model: core access is still licensed per user, but variable AI work is now represented by credits that can be topped up separately. Monthly customers also face a 17–25% seat-price increase while annual rates remain unchanged.
Google is changing Gemini Notebook’s packaging from feature-style quotas toward a compute budget. That gives users more flexibility but makes the effective cost of one request less predictable and ties premium upgrades more directly to computational intensity.