AI API pricing snapshot — September 2026: verified list prices across the frontier

Verification note (accessed 2026-09-05): Every price, context-window and release-date figure below is verified against two independent sources listed in Sources and against the site's model catalog, where each row carries its own check date. The August edition of this snapshot has been superseded by the late-August/September release wave: GPT-6 Astra, Gemini 3.8 Flash, Muse Spark 1.3 and Claude Fable 5.1 all shipped inside one week.

If you only read one paragraph: as of 2026-09-05 the frontier API tiers sit at GPT-6 Astra $10/$50 (OpenAI's new computer-use flagship), GPT-5.6 Sol $4/$20, Claude Fable 5.1 / Claude Opus 5 $10/$50 and $5/$25, Gemini 3.8 Flash $0.75/$3.75 (intro), Muse Spark 1.3 $1.25/$4.25, Grok 4.6 $2/$6, with strong open-weight budget options from DeepSeek V4 Pro ($1.32/$3.96 peak), GLM 5.3 Flash ($0.15/$0.50 third-party) and Mistral Large 3 ($0.50/$1.50). None of this is a forecast — it is the live list, and every catalog row shows the date it was last checked.

Verified list prices (per 1M tokens, standard tier)

Why this matters for buyers

• The frontier repriced twice in five weeks. GPT-5.6 Sol's 2026-08-21 cut was followed by four new flagships in the 2026-09-01 → 09-03 window. Any forecast built on the August snapshot is already one generation behind — check the model table, where every row carries its check date.
• One model no longer has one price. Surface, region, cache, and provider all change the number; the cheapest tier almost always carries conditions (Gemini 3.7/3.8 Flash intro pricing doubles on 2027-01-01 — verify the route before locking an annual commitment).
• Route by workload. With GPT-5.6 Luna at $0.20 and Gemini 3.8 Flash at $0.75 intro, per-task model routing beats single-model standardization for high-volume work. The model picker applies this logic to your answers.
• Open weights change the math. Mistral Large 3, DeepSeek V4 and GLM 5.3 weights are free to self-host, removing per-token API lock-in for capable tiers — the trade is serving ops.

Where the new releases land on price

• GPT-6 Astra ($10/$50) is the most expensive OpenAI tier — a computer-use flagship, not a Luna replacement. Long-context work above 263K tokens bills $20/$100.
• Gemini 3.8 Flash ($0.75/$3.75 intro) undercuts nothing in the table on capability per dollar if Google's Terminal-Bench numbers hold; the price doubles on 2027-01-01.
• Muse Spark 1.3 ($1.25/$4.25) undercuts every closed frontier row — with two caveats worth reading: the headline benchmark used a reasoning mode that is not broadly available yet, and the cheaper "contributor" tier trains on your data.

Related reading on the site

• GPT-6 Astra launch guide — what the new flagship changes.
• Gemini 3.8 Flash launch guide — the value story of the September wave.
• AI API pricing cliffs in September 2026 — the deeper cliff-by-cliff breakdown.
• Claude Opus vs budget flash for API spend — how to match tier to workload.

Sources (accessed 2026-09-05 unless noted)

• OpenAI — GPT-6 Astra ($10/$50, >263K $20/$100): https://openai.com/index/gpt-6-astra/
• OpenAI — GPT-5.6 (Sol $4/$20 promo; Terra $2/$12; Luna $0.20/$1.20): https://openai.com/index/gpt-5-6/
• Anthropic — Claude Fable 5.1 and Mythos 5.1 ($10/$50; cache reads $0.25): https://www.anthropic.com/claude-fable-and-mythos-5-1
• Anthropic — Claude Haiku 4.5 deprecation floor: https://platform.claude.com/docs/en/about-claude/model-deprecations
• Google — Gemini 3.8 Flash and 3.8 Flash Cyber (intro pricing, HLE-Verified 54.9): https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/
• Google — Gemini API pricing: https://ai.google.dev/gemini-api/docs/pricing
• Meta — Introducing Muse Spark 1.3 ($1.25/$4.25, cached $0.15): https://research.meta.ai/blog/introducing-muse-spark-1-3
• Meta developer — Muse Spark pricing table: https://developer.meta.com/ai/models/muse-spark/
• xAI — Grok 4.6 pricing ($2/$6, 500K): https://docs.x.ai/docs/pricing
• DeepSeek API pricing (V4 Pro $1.32/$3.96 peak, off-peak $0.66/$1.98): https://api-docs.deepseek.com/quick_start/pricing/
• PricePerToken — Mistral Large 3 2512 ($0.50/$1.50, 262K): https://pricepertoken.com/pricing-page/model/mistral-ai-mistral-large-2512

This post is research-derived and editorial. It does not replace a vendor quote or a committed-use contract. Verify live prices before purchasing or publishing a forecast.