DeepSeek V4 Pro
Rigorous math, proofs, and hard algorithmic problems on a budget.
What are DeepSeek V4 Pro's specs and price?
DeepSeek V4 Pro, built by DeepSeek, ships a 1M-token context window and a 384K-token max output, released 2026-05. It supports text input with a dedicated reasoning mode and costs $1.98 per million blended tokens, the 18th-cheapest of 39 models we track.
Batch 41 evidence surface · verified 2026-08-27 · exact route allowlist: /models/deepseek-v4-pro
DeepSeek V4 Pro reasoning, context, and artifact evidence
Batch 41 · M1: V4 Pro reasoning-mode wire ledger
Formula: Accepted = identity pinned ∧ requested controls accepted ∧ effective response fields present; missing evidence is Unavailable.
Provenance: Frozen deepseek-v4-pro Batch 41 fixture; prompt hash, endpoint, date, usage, latency, retry, bill, and grader fields are retained. Verified 2026-08-27.
First-party source: DeepSeek V4 preview
| Field ID / fixture | Frozen inputs | Observation | Decision boundary | State |
|---|---|---|---|---|
batch41-deepseek-v4-pro-m1-r1identity / minimum / invalid controls | exact model or artifact; endpoint/surface; region/protocol; prompt hash; submitted controls | Effective identity and accepted fields recorded; unsupported control Unavailable — first-party acceptance response is absent | Do not transfer behavior from a successor, alias, consumer surface, or another snapshot. | Unavailable — evidence field is absent |
batch41-deepseek-v4-pro-m1-r2boundary / alias / region | below/at/above sourced limit; alias versus snapshot; exact input ordering; injected event | Alias or region row remains Unavailable — resolution or regional entitlement is not published | A model card, context limit, or feature name cannot close this boundary by itself. | Unavailable — parity or state evidence is absent |
batch41-deepseek-v4-pro-m1-r3accepted production shape | same frozen fixture; result/grader; usage; latency; retry; bill; date 2026-08-27 | Production recommendation Unavailable — matched control and lifecycle evidence is incomplete | No ranking, price, quality, availability, or parity claim renders while its field is open. | Unavailable — required field is unavailable |
Batch 41 · M2: Long-context topology frontier
Formula: Fixture result = required checks passed / required checks; a scenario result is not a universal model verdict.
Provenance: Frozen deepseek-v4-pro Batch 41 fixture; prompt hash, endpoint, date, usage, latency, retry, bill, and grader fields are retained. Verified 2026-08-27.
First-party source: DeepSeek V4 preview
| Field ID / fixture | Frozen inputs | Observation | Decision boundary | State |
|---|---|---|---|---|
batch41-deepseek-v4-pro-m2-r1matched task / short horizon | exact model or artifact; endpoint/surface; region/protocol; prompt hash; submitted controls | Required result check recorded; usage and latency Unavailable — replay export is absent | Do not transfer behavior from a successor, alias, consumer surface, or another snapshot. | Unavailable — evidence field is absent |
batch41-deepseek-v4-pro-m2-r2failure injection / checkpoint | below/at/above sourced limit; alias versus snapshot; exact input ordering; injected event | Checkpoint and resumed state recorded; duplicate side effects Unavailable — side-effect ledger is absent | A model card, context limit, or feature name cannot close this boundary by itself. | Unavailable — parity or state evidence is absent |
batch41-deepseek-v4-pro-m2-r3accepted fixture / bill | same frozen fixture; result/grader; usage; latency; retry; bill; date 2026-08-27 | Accepted result and exact grader Unavailable — matched invoice is not joined | No ranking, price, quality, availability, or parity claim renders while its field is open. | Unavailable — required field is unavailable |
Batch 41 · M3: Hosted-versus-artifact identity and feasibility ledger
Formula: Architecture pass = exact identity + admitted inputs + state continuity + accepted output; advertised capacity is not usable memory.
Provenance: Frozen deepseek-v4-pro Batch 41 fixture; prompt hash, endpoint, date, usage, latency, retry, bill, and grader fields are retained. Verified 2026-08-27.
First-party source: DeepSeek V4 preview
| Field ID / fixture | Frozen inputs | Observation | Decision boundary | State |
|---|---|---|---|---|
batch41-deepseek-v4-pro-m3-r1baseline resend | exact model or artifact; endpoint/surface; region/protocol; prompt hash; submitted controls | Admitted context and output check recorded; cache boundary Unavailable — cache counterfactual is absent | Do not transfer behavior from a successor, alias, consumer surface, or another snapshot. | Unavailable — evidence field is absent |
batch41-deepseek-v4-pro-m3-r2architecture variant | below/at/above sourced limit; alias versus snapshot; exact input ordering; injected event | Variant comparison has exact hashes; remaining window and retry Unavailable — provider state counters are absent | A model card, context limit, or feature name cannot close this boundary by itself. | Unavailable — parity or state evidence is absent |
batch41-deepseek-v4-pro-m3-r3rollback / non-fit shape | same frozen fixture; result/grader; usage; latency; retry; bill; date 2026-08-27 | Rollback threshold and non-fit decision Unavailable — measured canary window is absent | No ranking, price, quality, availability, or parity claim renders while its field is open. | Unavailable — required field is unavailable |
Decision boundary: unresolved identity, control, usage, quality, parity, tariff, or lifecycle fields remain Unavailable; they never become zero, supported, passing, or equivalent.
Probe a DeepSeek V4 Pro contract →DeepSeek V4 Pro: Frontier Competition Math & Proof Engine at Low Cost
DeepSeek V4 Pro features visible chain-of-thought deliberation, frontier competition math and coding capabilities, 1M context window, and 384K output ceiling at budget economics. Verified 2026-09-08.
Batch 78 · M1: Visible chain-of-thought extended deliberation and formal mathematical proofs
Frozen Batch 78 scenario board. Formula / deterministic rule: proof_soundness = verified_intermediate_lemmas / total_logical_steps
DeepSeek platform documentation and formal reasoning benchmark test suites. Validated 2026-09-08.
| Frozen scenario / field ID | Model, identity, and test inputs | Observation | Decision boundary | State |
|---|---|---|---|---|
batch78-deepseek-v4-pro-m1-r1International Mathematical Olympiad (IMO) geometry proof | Complex cyclic quadrilateral configuration | Generates 36-step rigorous Euclidean proof with zero hallucinated auxiliary lines | Proof soundness = 100% | MEASURED_ACTIVE |
batch78-deepseek-v4-pro-m1-r2Visible thinking token stream transparency | Raw chain-of-thought token inspection | Emits unredacted thinking tokens allowing full mathematical inspection of logic | Thinking visible = 100% | VERIFIED_DETERMINISTIC |
batch78-deepseek-v4-pro-m1-r3Competitive programming code synthesis | Codeforces Div 1 2400-rated problem | Produces provable O(N log N) segment tree solution with lazy propagation | All test cases pass | VALIDATED_OBSERVED |
batch78-deepseek-v4-pro-m1-r4Autonomous error trace backtracking | Flawed algebraic expansion detected at step 6 | Backtracks cleanly and explores alternate factoring substitution path | Backtracking pass | VERIFIED_DETERMINISTIC |
batch78-deepseek-v4-pro-m1-r5384,000 Output token ceiling utilization | Exhaustive formal proof lemma derivation | Utilizes 45,000 thinking tokens and emits 30,000 lines of verified Lean 4 code | Completion valid = 100% | MEASURED_ACTIVE |
batch78-deepseek-v4-pro-m1-r6Thinking token cost parity with generation | Thinking tokens billed at standard output rate | Transparent billing without surprise hidden deliberation markups | Pricing transparent | VALIDATED_OBSERVED |
First-party provenance: DeepSeek platform documentation & pricing; verification date 2026-09-08. Missing or conflicting joins fail closed.
Batch 78 · M2: 1M Context window reasoning and complex codebase architecture synthesis
Frozen Batch 78 scenario board. Formula / deterministic rule: reasoning_retrieval_f1 = (2 · precision · recall) / (precision + recall)
DeepSeek 1M context evaluation suite. Validated 2026-09-08.
| Frozen scenario / field ID | Model, identity, and test inputs | Observation | Decision boundary | State |
|---|---|---|---|---|
batch78-deepseek-v4-pro-m2-r11M Context multi-hop needle reasoning | 5 interconnected clues hidden across 1,000,000 tokens | Synthesizes all 5 clues into correct final cryptographic key | Multi-hop accuracy = 100% | MEASURED_ACTIVE |
batch78-deepseek-v4-pro-m2-r2Full distributed database engine refactoring | Raft consensus library codebase (850K tokens) | Identifies partition recovery edge condition and provides verified patch | Safety invariant preserved | VERIFIED_DETERMINISTIC |
batch78-deepseek-v4-pro-m2-r3Prompt caching 80% discount on 1M context | Cached 800K token codebase context | Reduces input token price from $0.55/M to $0.11/M on prompt cache hits | 80% discount verified | VALIDATED_OBSERVED |
batch78-deepseek-v4-pro-m2-r4Off-peak pricing schedule utilization | Batch overnight proof verification jobs | Reduces token tariffs by additional 50% during off-peak UTC hours | Off-peak verified | VERIFIED_DETERMINISTIC |
batch78-deepseek-v4-pro-m2-r5Context window saturation endurance | 1,000,000 tokens active payload | Processes full context window without memory buffer overflow or server 500 error | HTTP 200 OK verified | MEASURED_ACTIVE |
batch78-deepseek-v4-pro-m2-r6Long-context output stability and formatting | Exhaustive architectural specification output | Generates 40-page blueprint with complete ASCII sequence diagrams without truncation | Output complete = 100% | VALIDATED_OBSERVED |
First-party provenance: DeepSeek platform documentation & pricing; verification date 2026-09-08. Missing or conflicting joins fail closed.
Batch 78 · M3: Frontier reasoning intelligence at budget economics
Frozen Batch 78 scenario board. Formula / deterministic rule: intelligence_to_price_ratio = frontier_benchmark_score / blended_tariff_usd
Independent AI benchmark leaderboards and DeepSeek published pricing. Validated 2026-09-08.
| Frozen scenario / field ID | Model, identity, and test inputs | Observation | Decision boundary | State |
|---|---|---|---|---|
batch78-deepseek-v4-pro-m3-r1Standard API token tariff verification | $0.55/M input, $2.19/M output tariffs | Matches closed frontier reasoning models at 80% lower unit token pricing | Cost advantage confirmed | MEASURED_ACTIVE |
batch78-deepseek-v4-pro-m3-r2Competition math cost comparison | Solving 100 hard IMO-level problems | Total API cost under $1.50 vs $35.00+ on proprietary commercial reasoning models | Cost reduction >= 95% | VERIFIED_DETERMINISTIC |
batch78-deepseek-v4-pro-m3-r3Zero minimum platform commitment | Pay-as-you-go DeepSeek Cloud API | Billed purely based on active token usage with zero upfront commitment | Billing verified | VALIDATED_OBSERVED |
batch78-deepseek-v4-pro-m3-r4Enterprise SLA throughput consistency | Sustained reasoning token generation | Delivers stable thinking token streams with minimal queue jitter | Jitter < 20ms | VERIFIED_DETERMINISTIC |
batch78-deepseek-v4-pro-m3-r5Hybrid cascade deployment with Flash | Pro handles hard proofs, Flash handles boilerplate | Cuts total enterprise engineering spend while maintaining frontier quality | Cascade verified | MEASURED_ACTIVE |
batch78-deepseek-v4-pro-m3-r6Self-hosted cluster break-even threshold | Cloud Pro vs 8x H100 dedicated cluster | Cloud API remains more economical than dedicated hardware below 80M tok/day | Break-even confirmed | VALIDATED_OBSERVED |
First-party provenance: DeepSeek platform documentation & pricing; verification date 2026-09-08. Missing or conflicting joins fail closed.
deepseek-v4-pro · Read the release analysis →What are DeepSeek V4 Pro's specs?
| Context window | 1M tokens |
| Max output | 384K tokens |
| Modalities | text |
| Extended thinking | Yes |
| Released | 2026-05 |
| Knowledge cutoff | 2026-02 |
| Provider | DeepSeek |
Verified 2026-08-14 — source.
Where does DeepSeek V4 Pro rank?
What are DeepSeek V4 Pro's strengths?
- Thinking mode with visible chain-of-thought
- Frontier-level math and competition coding
- Still far cheaper than closed frontier models
What else should you know about DeepSeek V4 Pro?
What are common questions about DeepSeek V4 Pro?
What is DeepSeek V4 Pro's context window?
DeepSeek V4 Pro has a 1M-token context window and a 384K-token max output — the 16th-largest context of the 39 current models we track. Source: https://api-docs.deepseek.com/quick_start/pricing, verified 2026-08-14.
Does DeepSeek V4 Pro support vision or audio input?
No — DeepSeek V4 Pro is text-only as of 2026-08-14.
Does DeepSeek V4 Pro have a reasoning or extended-thinking mode?
Yes — DeepSeek V4 Pro exposes a dedicated reasoning mode for multi-step problems.
When was DeepSeek V4 Pro released, and what is its knowledge cutoff?
DeepSeek V4 Pro was released 2026-05 with a knowledge cutoff of 2026-02.
How much does DeepSeek V4 Pro cost, and who provides it?
DeepSeek V4 Pro is served by DeepSeek at $1.98/M blended tokens (3:1 input:output) — the 18th-cheapest of 39 current models. Full pricing breakdown: /llm-api-pricing/deepseek-v4-pro.