What are the official 2026 LLM API prices per 1M tokens for GPT-6.1 Sol, Claude Opus 5.5, Sonnet 5.5, Gemini 4 Argon, and DeepSeek V4 Pro?

Verified v17-unified rates (per 1M tokens, input / cached input / output): OpenAI GPT-6.1 Sol is $2.00 / $0.10 / $10.00 (>272K cliff: $4.00 / $0.20 / $15.00); Claude Opus 5.5 is $4.00 / $0.20 / $20.00 (1M flat window); Claude Sonnet 5.5 is $2.00 / $0.10 / $10.00 (1M flat window); Claude Haiku 5.5 is $0.10 / $0.01 / $0.50 (>100K 5x cliff: $0.50 / $0.05 / $2.50); Gemini 4 Argon is $2.00 / $0.10 / $10.00 through Nov 30, 2026 ($4.00 / $0.20 / $20.00 from Dec 1, 2026); DeepSeek V4 Pro is $0.66 / $0.022 / $1.98 Off-Peak ($1.32 / $0.044 / $3.96 Peak); DeepSeek V4.1-Flash-1017 is $0.15 / $0.003 / $0.60 Off-Peak.

How does Etuki calculate Effective Cost per Successful Task (C_success) and verify real customer bills via Bill X-Ray & Cost Regression CI?

Etuki computes C_success = E[C_total] / P(success) via /api/v1/effective-cost, calibrating for tokenizer density (Claude 4.7+/5.5 ~1.28x vs OpenAI o200k 1.00x), piecewise context cliffs, prefix cache hits, and transport retries. Teams can verify real CSV bills in the browser-local Bill X-Ray (#live-bill-verifier), replay REG-01..REG-05 in /api/v1/gateway-sandbox across 4 rollout stages (Dry-Run, Shadow, Canary 10%, Production + Auto-Rollback), and block >15% PR cost regressions via /api/v1/ci-gate.

How does the Etuki Universal FinOps Execution Compiler prevent 272K, 200K, 100K, and 256K context-cliff bill spikes?

Etuki enforces a two-tier soft-block routing architecture: Tier 1 intercepts prompts at 90% of hard vendor cliffs (244,800 tokens for OpenAI >272K, 180,000 for Grok/Gemini >=200K, 90,000 for Haiku >100K) and reroutes overflow to flat-window Claude Sonnet 5.5. Tier 2 takes precedence for Alibaba Qwen 3.7 Flash, keeping 32K-230.4K prompts on Qwen 3.7 Flash (0.96 CNY/1M) and rerouting >230,400 tokens (90% of 256K cliff) to Qwen 3.8 Flash (0-1M flat tier at 1.18 CNY/1M).

✅ SCHEMA 2026.10.30-v39.1-unified • SNAPSHOT price-20261030-v35-truth • RS256 SIGNED MANIFEST

Etuki AI Cost Decision Infrastructure — Full-Stack Verification Ledger (v39.1)

Unified cryptographic trust chain binding /api/oracle/v1.json, /api/oracle/leaves-563.json, /api/telemetry/metrics.json, /api/v1/mcp, and cooyun/etuki-gate.

🔐 P0.1: Cryptographically Signed Unified Evidence Manifest & Three Merkle Roots

schema_version: 2026.10.30-v39.1-unified

pricing_snapshot_id: price-20261030-v35-truth

telemetry_snapshot_id: telemetry-20261030-v35-truth

1. catalog_root_sha256 (Binary Tree): sha256:d2f517094b814bcf41ea821c1ce654aa29df8ac454acd1c96779d7966ab688f4

2. dataset_3branch_csv_root_sha256: sha256:c2408e0410fc4243bdb2f0083605d4fd5fbb322b94cca7a94359c4eb0b1aa79e

3. price_payload_sha256 (563 Leaves): sha256:6bc877254427a142ff842bfa318350453ffebc62faf90f62337f96fb7fa5db82

563 Leaves Public Endpoint: /api/oracle/leaves-563.json

signature_key_id: etuki-ci-rsa2048-2026-v17 (RS256)

canonicalization: PYTHON-SORTED-KEYS-COMPACT-UTF8-V1

signed_payload_sha256: f92a7b89f27028500d22ba74f4dd9c9805d1af24499f82eae7b7c96c6a8b501b

signature_value (512 hex): 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

measured_at / stale_after: 2026-10-30T02:10:00Z / 2026-10-31T03:05:00Z

🏛️ P2: E1 / E2 / E3 Evidence Certification & Revocation Registry (v39.1)

Certificate ID	Evidence Tier	Issued At	Status	Strict Claim Boundary
ETK-E1-20261030-SNAP35	E1: Data Verified	2026-10-30T02:10:00Z	ACTIVE_VALID	Rate-card snapshot & Merkle inclusion only; not customer bill savings.
ETK-E2-20261030-BENCH12	E2: Benchmark Verified	2026-10-30T02:10:00Z	ACTIVE_VALID	Reproducible 12-vector CLI/MCP math under synthetic workload only.
ETK-E3-20261029-PILOT01	E3: Economics Verified	N/A (Unissued)	DEMO_PILOT_TEMPLATE_UNISSUED	Unissued template: Zero active E3 certificates claimed today.
ETK-E3-20260915-REVOKED01	E3: Economics Verified	2026-09-15T09:00:00Z	REVOKED	Revoked: Superseded by Oct 2026 rate-card snapshot.

Date

Model (Official API ID)

In / Out (1M)

Official Cache Read

Context Cliff / Repricing Rule

2026-10-17

DeepSeek-V4.1-Flash-1017 (deepseek-v4.1-flash-1017)

$0.15 / $0.60

$0.003/1M (98% Off)

[snap-20261017-v1] DeepSWE 77.0% • Search/Fetch API $0.75/1K (from Oct 31 16:00Z)

2026-10-07

Claude Haiku 5.5 (claude-haiku-5-5)

$0.10 / $0.50

$0.01/1M (≤100K, 90% Off)

[snap-20261007-b] >100K Cliff jumps 5x to $0.50/$2.50 (Cache $0.05)

2026-10-07

Claude Sonnet 5.5 (claude-sonnet-5-5)

$2.00 / $10.00

$0.10/1M (95% Off)

[snap-20261007-a] Official cache read cut 50% from $0.20 → $0.10/1M

2026-10-01

Gemini 4 Argon (gemini-4-argon)

$2.00 / $10.00

$0.10/1M (95% Off)

[snap-20261001-v1] Intro ends 2026-11-30 → $4.00/$20.00 (Cache $0.20) on 12/01

2026-09-29

GPT-6.1 Sol (gpt-6.1-sol / gpt-6-1-sol)

$2.00 / $10.00

$0.10/1M (95% Off)

[snap-20260929-v1] >272K Context Cliff jumps to $4.00/$15.00 (Cache $0.20)

2026-09-28

Claude Sonnet 5.5 GA (claude-sonnet-5-5)

$2.00 / $10.00

$0.20/1M (Initial GA)

[snap-20260928-v1] Initial GA launch (later cut to $0.10 cache on Oct 7)

2026-09-22

Claude Opus 5.5 (claude-opus-5-5)

$4.00 / $20.00

$0.20/1M (95% Off)

[snap-20260922-b] 20% cheaper than Opus 5 ($5/$25 → $4/$20)

2026-09-22

GPT-6 Luna (gpt-6-luna)

$0.10 / $0.50

$0.01/1M (90% Off)

[snap-20260922-a] 100x cheaper than Astra • >272K Cliff $0.20/$0.75

2026-09-03

GPT-6 Astra (gpt-6-astra)

$10.00 / $50.00

$1.00/1M (90% Off)

[snap-20260903-v1] 90% Official Cached Input • >272K Cliff $20/$75

2026-09-01

Claude Fable 5.1 (claude-fable-5-1)

$10.00 / $50.00

$0.25/1M (97.5% Off)

[snap-20260901-v1] Frontier research launch with 97.5% cache read discount

2026-08-13

DeepSeek-V4 Pro (deepseek-v4-pro)

$0.66 / $1.98

$0.022/1M (96.7% Off)

[snap-20260813-v1] GA Launch • Peak 2x ($1.32/$3.96, eff. 08-17) / Off-Peak 50% Off

2026-07-31

DeepSeek-V4 Flash Beta (deepseek-v4-flash-0731-beta)

$0.15 / $0.60

$0.003/1M (98% Off)

[snap-20260731-v1] Public Beta Launch (upgraded by V4.1-Flash-1017 on Oct 17)

Policy Tier & Cliff

Soft-Block Trigger

Test Input

Source Model -> Fallback Target

Regression Verdict

REG-01: Tier 1 (>272K 2x Cliff)

244,800 soft-block (90% guard)

250K guard / 280K+10K over-cliff

openai/gpt-6.1-sol -> anthropic/claude-sonnet-5.5

PASS (Preventive @244.8K; @280K saves $1.27 vs $0.66)

REG-02: Tier 1 (>=200K 2x Cliff)

180,000 soft-block (90% guard)

185K guard / 210K+8K over-cliff

x-ai/grok-4.7 -> anthropic/claude-sonnet-5.5

PASS (Preventive @180K; @210K saves $0.936 vs $0.500)

REG-03: Tier 1 (>100K 5x Cliff)

90,000 soft-block (90% guard)

95K guard / 105K+2K over-cliff

anthropic/claude-haiku-5.5 -> openai/gpt-6-luna (or Sonnet 1M)

PASS (@105K 5x cliff $0.0575 -> $0.0115 on Luna)

REG-04: Tier 2 Precedence (>32K)

28,800 step soft-cap

30K guard / 35K+2K step

qwen/qwen-3.7-flash (<=32K) -> stays on qwen-3.7-flash (32K-256K)

PASS (0.96 CNY/1M vs 1.18 CNY/1M; never Sonnet)

REG-05: Tier 2 (>256K Crossover)

230,400 soft-cap (90% of 256K)

240K guard / 260K+10K over-cliff

qwen/qwen-3.7-flash -> qwen/qwen-3.8-flash (0-1M flat)

PASS (@260K/10K saves 34.7% [0.235 vs 0.360 CNY]; 80/20 blended saves 38.5% [1.18 vs 1.92 CNY/1M])