AI API pricing snapshot — October 2026: current frontier and budget tiers

API prices and status checked on 1 October 2026. Rows that retain earlier price checks say so explicitly; rates for legacy or previous models do not prove availability on every account or endpoint. Gemini 4 Argon is a restricted preview and has no public purchasable API price verified in the cited announcement.

If you only read one paragraph: the newest OpenAI Sol tier, GPT-6.1 Sol, is $2/$10 per million input/output tokens (cached input $0.10); Claude Sonnet 5.5 is $2/$10 (cached input $0.20). Other current API rows include GPT-6 Astra $10/$50, Claude Fable 5.1 $10/$50 (check eligibility), Claude Opus 5.5 $4/$20, Mistral Medium 3.5 $1.50/$7.50, Grok 4.7 $2/$6, Muse Spark 1.3 $1.25/$4.25 and Gemini 3.8 Flash $0.75/$3.75 (intro). The lower-priced end includes GPT-6 Luna $0.10/$0.50 and DeepSeek V4.1 Flash $0.30/$1.20 at peak ($0.15/$0.60 off-peak). The table is a point-in-time price snapshot: rows for previous/legacy models preserve dated source checks and do not imply every model remains selectable on every account or endpoint. Rates have conditions and are not a universal task-cost ranking.

Published API rates and dated comparison rows (per 1M tokens, standard tier)

Context windows come from each provider's model docs. Check them again before sizing long-context work, because several models bill long prompts at a higher rate.

What changed in September

• OpenAI lowered Sol and Luna prices. GPT-6 Sol ($2/$10) halves GPT-5.6 Sol's promotional $4/$20. GPT-6 Luna ($0.10/$0.50) halves its predecessor's input price and lowers output from $1.20 to $0.50 (about 58%). GPT-6 Sol now costs less on output than GPT-5.6 Terra. The migration steps are in Migrating from GPT-5.6 to GPT-6 Sol and Luna.
• OpenAI released GPT-6.1 Sol on 29 September. It keeps the $2/$10 standard rate, with cached input at $0.10/MTok instead of GPT-6 Sol's $0.20. This page lists both API IDs because the GPT-6 Sol API model page remains available separately from GPT-6.1 Sol.
• Anthropic released Sonnet 5.5 on 28 September. Its listed $2/$10 base token rates match Sonnet 5; the claimed efficiency improvement is a vendor-reported cost-per-task result, not a lower list rate.
• Anthropic cut Opus by 20% per token and 60% on cache reads. Opus 5.5 lists at $4/$20 with $0.20 cache reads. Anthropic says typical workloads cost about 40% less than on Opus 5, because the model also uses fewer tokens. Opus 5 moved to the prior-models group on Anthropic's pricing page.
• DeepSeek retired V4 Flash. V4.1 Flash replaced it on 10 September at $0.30/$1.20 peak. The old deepseek-v4-flash name is still accepted, but routes to V4.1 Flash and bills at the V4.1 Flash rate. Peak hours are 01:00–04:00 and 06:00–10:00 UTC Monday–Friday, except Chinese public holidays, which are off-peak for the full day.
• xAI shipped Grok 4.7 at Grok 4.6's standard API price. $2/$6 for standard prompts, with $4/$12 long-context rates at 200K tokens and above. xAI says it is available through Cursor, Grok Build, the API, third-party coding harnesses, model routers and cloud platforms.
• Mistral Medium 3.5 is a current API and open-weight option. Its API lists $1.50/$7.50, with $0.15 cached input per million tokens. The 128B dense checkpoint has 256K context and none/high reasoning. Modified MIT restricts licence rights when company/employer global consolidated revenue exceeded $20M in the preceding month. Docs/changelog record April 28; the legal centre lists April 29, and the May 22 article describes public preview. The current model card is GA. These are different dated source statements, not a verified single launch chronology.

Why this matters for buyers

• Frontier prices fell faster than the snapshot cycle. Any forecast built on the early-September edition is now a generation behind. Every row in the model table shows the date it was last checked.
• Cheapest per token is not necessarily cheapest per task. Anthropic reports that Opus 5.5 typically uses fewer tokens than Opus 5; that vendor result does not establish its relative cost against GPT-6 Sol. Compare cost per successful task on your own workload. GPT-6 Sol vs Claude Opus 5.5 walks through that calculation.
• One model no longer has one price. Surface, region, cache, time of day (DeepSeek) and provider all change the number. The cheapest tier usually carries conditions: Gemini 3.7 and 3.8 Flash intro pricing doubles on 2027-01-01, while OpenAI guarantees GPT-5.6 Sol's promo price at least through 2026-11-21 but has not announced an end date or replacement rate.
• Compare routing on your workload. GPT-6 Luna lists at $0.10/$0.50 and DeepSeek V4.1 Flash at $0.15/$0.60 off-peak. Those rates alone do not show which model will cost less per successful task; test representative tasks before standardising. The model picker helps narrow candidates.
• Open weights change the cost structure. Mistral Large 3 (Apache 2.0) and Medium 3.5 (Modified MIT) have open weights under their stated licences; self-hosting avoids a hosted per-token API charge but still has hardware, power and operations costs.

Related reading on the site

• GPT-6 Sol vs Claude Opus 5.5 — which to route to after the September price cuts.
• GPT-6 Astra launch guide — what the flagship changes.
• AI API pricing cliffs — the scheduled price steps worth diarising.
• Claude Opus vs budget flash for API spend — how to match tier to workload.

Sources (checked 2026-10-01 unless a row says otherwise)

• OpenAI — Introducing GPT-6 Sol and Luna ($2/$10, $0.10/$0.50; 90% cached-input discount): https://openai.com/index/introducing-gpt-6-sol-and-luna/
• OpenAI — GPT-6 Astra launch post: https://openai.com/index/gpt-6-astra/
• OpenAI — API pricing (GPT-6 Astra ≤272K $10/$50, >272K $20/$75; GPT-6 Sol >272K $4/$15; GPT-6 Luna >272K $0.20/$0.75): https://developers.openai.com/api/docs/pricing
• OpenAI — model docs for gpt-6-sol and gpt-6-luna (1.05M context, 128K max output): https://developers.openai.com/api/docs/models/gpt-6-sol
• OpenAI — GPT-5.6 (Sol $4/$20 promo; Terra $2/$12; Luna $0.20/$1.20), accessed 2026-09-05: https://openai.com/index/gpt-5-6/
• Anthropic — Introducing Claude Opus 5.5 ($4/$20; cache reads $0.20, writes $5; fast mode $8/$40): https://www.anthropic.com/claude-opus-5-5
• Anthropic — Introducing Claude Sonnet 5.5 (release date and same $2/$10 base rates as Sonnet 5): https://www.anthropic.com/claude-sonnet-5-5
• OpenAI — Introducing GPT-6.1 Sol (2026-09-29 launch): https://openai.com/index/introducing-gpt-6-1-sol/ — checked 2026-10-01
• OpenAI — GPT-6.1 Sol model documentation (price, cache and long-prompt conditions): https://developers.openai.com/api/docs/models/gpt-6.1-sol — checked 2026-10-01
• OpenAI — GPT-6 Sol, GPT-6 Luna and GPT-5.6 Terra/Luna API model documentation: https://developers.openai.com/api/docs/models/gpt-6-sol ; https://developers.openai.com/api/docs/models/gpt-6-luna ; https://developers.openai.com/api/docs/models/gpt-5.6-terra ; https://developers.openai.com/api/docs/models/gpt-5.6-luna — checked 2026-10-01
• Anthropic — Claude API pricing (Fable 5.1, Opus 5.5, Sonnet 5.5, Haiku 4.5 and prior-model references): https://platform.claude.com/docs/en/about-claude/pricing — checked 2026-10-01
• Anthropic — model deprecations (Haiku 4.5 support floor): https://docs.anthropic.com/en/docs/about-claude/model-deprecations — checked 2026-10-01
• The Batch (DeepLearning.AI) — Claude Opus 5.5 (1M-token input context): https://www.deeplearning.ai/the-batch/claude-opus-5-5-leaps-forward
• Google — Gemini API pricing for Gemini 3.7/3.8 Flash, including Jan 1 2027 rate step: https://ai.google.dev/gemini-api/docs/pricing — checked 2026-10-01
• Google — Gemini API model list and deprecations: https://ai.google.dev/gemini-api/docs/models ; https://ai.google.dev/gemini-api/docs/deprecations — checked 2026-10-01
• Meta — Muse Spark pricing and tier-specific data-use terms: https://dev.meta.ai/docs/pricing-rate-limits — checked 2026-10-01
• xAI — Introducing Grok 4.7 (2026-09-21 launch, price and stated distribution): https://x.ai/news/grok-4-7 — checked 2026-10-01
• xAI — API pricing for Grok 4.7 and 4.6, including long-context rates: https://docs.x.ai/developers/pricing — checked 2026-10-01
• DeepSeek — Models & Pricing (V4.1 Flash $0.30/$1.20 peak, off-peak half; V4 Pro $1.32/$3.96; Chinese public holiday schedule): https://api-docs.deepseek.com/quick_start/pricing/ — checked 2026-10-01
• DeepSeek — API change log (V4.1 Flash release and V4 Flash retirement, 10 Sep 2026; V4 Pro continues): https://api-docs.deepseek.com/updates/
• Mistral — Large 3 model card (256K context and current $0.50/$1.50 rates): https://docs.mistral.ai/models/mistral-large-3-25-12
• Mistral — Large 3 launch and Apache 2.0 license: https://mistral.ai/news/mistral-3/ — accessed 2026-10-01
• Mistral — Medium 3.5 model details and current API price ($1.50/$7.50; 256K; Modified MIT): https://docs.mistral.ai/models/mistral-medium-3-5-26-04 — checked 2026-10-01
• Mistral — May 2026 Medium 3.5 launch and architecture (128B dense): https://mistral.ai/news/vibe-remote-agents-mistral-medium-3-5/

This snapshot is editorial research. It does not replace a vendor quote or a committed-use contract. Check live prices before you buy or publish a forecast.