Official docs · as of 2026-11-18

GPT-6.2 Astra vs Claude Fable 5.2: Both $8/$40 Compared (2026-11)

As of 2026-11-18, OpenAI's gpt-6.2-astra (released 2026-11-17) and Anthropic's claude-fable-5-2 (released 2026-11-09) meet at the exact same headline official API rate: $8 base input and $40 output per 1M tokens (each down 20% from its predecessor's $10 / $50). Beneath that headline, they diverge sharply on cache reads ($0.40 vs $0.20), long-context billing (>272K 2x/1.5x surcharge vs flat 1M pricing), AA-3D (1,318 vs 1,271 Elo), and Agents' Last Exam (+5.5 points for Astra). Both flagships are live on QCode, where one cr_ key powers Claude Code and Codex together.

Updated 2026-11-18

Switch models on the same endpoint and measure the cost per token yourself; check /models for callable models.

#Both $8 / $40 official rate#Cache read $0.40 vs $0.20#AA-3D 1,318 vs 1,271#1.05M (>272K surcharge) vs 1M

Four Key Differences Behind the Same $8 / $40 Price

$8 / $40

Identical base input and output official API prices

Per OpenAI's official API pricing (≤272K short-context Standard tier) and Anthropic's official API pricing, both gpt-6.2-astra and claude-fable-5-2 cost $8 base input and $40 output per 1M tokens (20% below gpt-6-astra and claude-fable-5-1 at $10 / $50); both also price Batch at $4 / $20 (50%) and Fast mode at $16 / $80 (2x).

$0.40 / $0.20

Prompt cache read and write rate gap

On OpenAI's official rate card, gpt-6.2-astra prices cached input reads at $0.40 (0.05x base input), 30-minute cache writes at $10, and 12-hour cache writes at $20; on Anthropic's official rate card, claude-fable-5-2 prices cache hits/refreshes at $0.20 (0.025x base input, half of Astra's rate), 5-minute cache writes at $10 (1.25x), and 1-hour cache writes at $16 (2x).

1.05M / 1M

Context window and >272K long-context billing rule

gpt-6.2-astra offers a 1,050,000 (1.05M) context window, 922,000 max input, and 128,000 max output, but when input exceeds 272K tokens the full request is billed at 2x input/cache ($16 / $0.80) and 1.5x output ($60); claude-fable-5-2 offers a 1,000,000 (1M) context window and 128,000 (128K) max output at flat $8 / $40 standard pricing across the full 1M window.

1,318 / 1,271

AA-3D, Agents' Last Exam & third-party indices

OpenAI's published benchmarks put gpt-6.2-astra at #1 on AA-3D with 1,318 Elo versus 1,271 Elo for claude-fable-5-2 (a 47-Elo lead, with gpt-6.2-sol at 1,259 Elo), and +5.5 points ahead of Fable 5.2 on Agents' Last Exam. On Artificial Analysis (2026-11-17), Astra scores 62 (#2) on the Intelligence Index while matching gpt-6-astra on the Coding Agent Index.

Same $8 / $40 Headline, Different Real-World Token Bills

As of 2026-11-18, although gpt-6.2-astra and claude-fable-5-2 share the same $8 input and $40 output per 1M tokens on their vendors' official pricing pages, multi-turn engineering workloads produce very different bills. First is cache-hit cost: in long agentic tool-use loops, claude-fable-5-2's official cache read rate is $0.20 / 1M tokens (0.025x), which is 50% lower than gpt-6.2-astra's $0.40 / 1M tokens (0.05x); for longer cache retention, OpenAI offers 30-minute ($10) and 12-hour ($20) writes, while Anthropic offers 5-minute ($10) and 1-hour ($16) writes. Second is the long-context threshold: while gpt-6.2-astra has a slightly larger total window (1,050,000 vs 1,000,000), any single request above 272K input tokens doubles input and cache rates and multiplies output by 1.5x ($16 input / $60 output) for the entire request; Anthropic's official pricing page states that claude-fable-5-2 includes its 1M window at flat standard rates, so a 900K-token prompt pays the same per-token rate as a 9K-token prompt. See /claude-fable-5-2-pricing and /gpt-6-2-pricing for full rate tables.

Benchmarks, Thinking Controls & Tooling Ecosystem Compared

Across reasoning benchmarks and API mechanics, each flagship has clear strengths: ① Benchmarks — per OpenAI's 2026-11-17 release data, gpt-6.2-astra leads AA-3D at 1,318 Elo (vs 1,271 Elo for claude-fable-5-2), beats Fable 5.2 by 5.5 points and GPT-6 Astra by 5.2 points on Agents' Last Exam (at less than one-third the task cost of GPT-6 Astra), scores 97.0% on GPQA Diamond (no tools), and reaches 53.5% code execution on ExploitBench Internal Port (up 22 percentage points from 31.5% on GPT-6 Astra, with Critical Cybersecurity and High AI Self-Improvement ratings on the system card); independent evaluator Artificial Analysis (2026-11-17) places gpt-6.2-astra at 62 (#2) on its Intelligence Index with a Coding Agent Index matching gpt-6-astra. ② Thinking & parameter rules — gpt-6.2-astra supports low, medium, high, xhigh, and max reasoning effort but explicitly rejects reasoning_effort="none"; claude-fable-5-2 keeps Adaptive thinking always on (default effort high), requires omitting temperature, top_p, and top_k, and returns a 400 error on forced tool use (tool_choice set to any or tool).

Flagship Release Timeline

2026-11-09

Anthropic releases Claude Fable 5.2 (claude-fable-5-2) with a 1M context window, 128K max output, and official API pricing of $8 input / $40 output / $0.20 cache read; Claude Code 2.1.321 makes it the default Fable model on the same day.

2026-11-17

OpenAI releases GPT-6.2 Astra (gpt-6.2-astra) and GPT-6.2 Sol (gpt-6.2-sol) with a 1,050,000 context window, 128,000 max output, and matching official API pricing for Astra at $8 input / $40 output ($0.40 cached input).

2026-11-18

QCode supports both claude-fable-5-2 and gpt-6.2-astra in production, letting developers run Claude Code and Codex from one cr_ API key; all figures on this page were verified on 2026-11-18.

Confirmed Details vs Unannounced Items

Confirmed (OpenAI & Anthropic official documentation)

① Release dates & cutoffs: gpt-6.2-astra shipped on 2026-11-17 with a 2026-05-18 knowledge cutoff; claude-fable-5-2 shipped on 2026-11-09. ② Context & modalities: Astra has a 1,050,000 context window, 922,000 max input, and 128,000 max output; Fable 5.2 has a 1,000,000 context window and 128,000 max output (1M enabled by default with no beta header); both accept text and image input and produce text output. ③ Official vendor API pricing (per 1M tokens): both charge $8 base input and $40 output, $4 / $20 Batch, and $16 / $80 Fast; Astra also offers Ultrafast (3x or $24 / $120 through 2027-02-28, 6x from 2027-03-01); cache reads are $0.40 on Astra vs $0.20 on Fable 5.2; above 272K input tokens Astra charges 2x input/cache and 1.5x output for the full request, while Fable 5.2 has no long-context surcharge across 1M. ④ Reasoning & tools: Astra rejects reasoning_effort="none" and supports Multi-agent beta, computer_use, apply_patch, hosted_shell, mcp, and tool_search in the Responses API; Fable 5.2 has always-on adaptive thinking and requires omitting sampling temperature parameters.

Not published or important caveats

As of 2026-11-18, neither OpenAI nor Anthropic has published an official same-source SWE-bench Verified head-to-head table for these two models. Also note that Claude Mythos 5.2, while sharing Fable 5.2's specs and price on Anthropic's overview page, is restricted to organizations approved through Anthropic's verification programs; on OpenAI's side, starting December 2026, Plus, Pro, and Business usage allowances will be shared between ChatGPT and Codex (the exact day in December 2026 has not yet been announced).

How to Split Workloads Between the Two $8 / $40 Flagships

Multi-agent Codex workflows, 3D/UI & hardest reasoning: pick gpt-6.2-astra

When your task relies on Responses API Multi-agent beta subagent delegation, native hosted_shell and apply_patch tools, 3D/spatial generation (#1 on AA-3D at 1,318 Elo), or deep multi-step reasoning (+5.5 points over Fable 5.2 on Agents' Last Exam and 97.0% on GPQA Diamond), run gpt-6.2-astra in Codex CLI (paired with $2 / $10 gpt-6.2-sol for routine subtasks) while keeping individual prompts at or below 272K tokens.

Claude Code sessions with >272K repo context & heavy caching: pick claude-fable-5-2

When your team lives in Claude Code (v2.1.321+) or regularly feeds 300K–900K tokens of repository context into a single request, claude-fable-5-2 avoids OpenAI's >272K 2x/1.5x surcharge by keeping flat $8 / $40 pricing across the full 1M window while cutting official cache-read costs in half ($0.20 vs $0.40 per 1M tokens).

Configure Both Claude Code and Codex with One QCode Key

A single QCode cr_ API key works across both Anthropic Messages and OpenAI Responses / Chat protocols, routed automatically by endpoint path: ① Run claude-fable-5-2 in Claude Code (v2.1.321 or later): set ANTHROPIC_BASE_URL to https://api.qcode.cc/api and ANTHROPIC_AUTH_TOKEN to your cr_ key, then type /model fable or /model claude-fable-5-2 in your session (or set "model": "claude-fable-5-2" in ~/.claude/settings.json). ② Run gpt-6.2-astra in Codex CLI: in ~/.codex/config.toml set model = "gpt-6.2-astra", model_provider = "qcode", and model_reasoning_effort = "high" (or "xhigh", never "none"); under [model_providers.qcode] set base_url = "https://api.qcode.cc/openai" (or https://asia.qcode.cc/openai in mainland China), wire_api = "responses", and requires_openai_auth = true, and place the same cr_ key in OPENAI_API_KEY inside ~/.codex/auth.json. ③ Run gpt-6.2-astra in OpenAI SDK / Cline / OpenCode: set base_url to https://api.qcode.cc/openai/v1.

One Shared Token Balance for Both Flagships on QCode

Instead of paying for two separate monthly subscriptions, QCode supports gpt-6.2-astra, gpt-6.2-sol, and claude-fable-5-2 (plus claude-opus-5-5, qwen4-max, and more) from one token-billed balance; see live per-model rates on /models. Not ready for a monthly plan? Pay-as-you-go Packs start at $4 for a one-time quota you use within 7 days; if you already have a Personal Plan, add a booster pack when you need more quota.

Frequently Asked Questions

Are gpt-6.2-astra and claude-fable-5-2 priced identically on their official APIs?

For short-context requests (≤272K input tokens) at Standard tier, both vendors' official API pricing pages list $8 base input and $40 output per 1M tokens (with Batch at $4 / $20 and Fast at $16 / $80). They differ on caching and long context: claude-fable-5-2 charges $0.20 for cache reads ($16 for 1-hour writes) with flat pricing across its full 1M window; gpt-6.2-astra charges $0.40 for cache reads ($10 for 30-minute writes, $20 for 12-hour) and bills the full request at $16 input / $60 output when input exceeds 272K tokens. QCode bills per token, with per-model rates on /models.

Which model is more cost-effective for prompts above 272K input tokens?

Under each vendor's official billing rules, a prompt between 272K and 922K input tokens on gpt-6.2-astra triggers the 2x input/cache ($16 / $0.80) and 1.5x output ($60) surcharge for the entire request; Anthropic's official pricing page keeps claude-fable-5-2 at the standard $8 input / $40 output / $0.20 cache read across the entire 1,000,000-token context window.

How do they compare on AA-3D, Agents' Last Exam, and Artificial Analysis?

Per OpenAI's 2026-11-17 release benchmarks, gpt-6.2-astra ranks #1 on AA-3D at 1,318 Elo vs 1,271 Elo for claude-fable-5-2, and scores 5.5 points higher than claude-fable-5-2 on Agents' Last Exam. On Artificial Analysis (2026-11-17), gpt-6.2-astra scores 62 (#2) on the Intelligence Index, while its Coding Agent Index matches the previous-generation gpt-6-astra.

Can I call gpt-6.2-astra directly inside Claude Code?

No. Per QCode's endpoint specification, Claude models use the Anthropic protocol endpoint https://api.qcode.cc/api, whereas GPT models use the OpenAI protocol endpoints. However, the same QCode cr_ API key works in both Claude Code (pointed at https://api.qcode.cc/api for claude-fable-5-2) and Codex CLI (pointed at https://api.qcode.cc/openai for gpt-6.2-astra) simultaneously from one account balance.

What parameter gotchas should I watch out for when switching between them?

gpt-6.2-astra rejects API requests with reasoning_effort="none" (use low, medium, high, xhigh, or max instead); claude-fable-5-2 keeps Adaptive thinking always on (controlled via effort, default high), requires omitting temperature, top_p, and top_k, and returns a 400 error if tool_choice forces tool use (any or tool).

What are the lower-cost companion models for everyday coding in each family?

On the OpenAI side, pair Astra with gpt-6.2-sol ($2 / $10 official API price, within 0.5 points of Astra on FrontierCode 2.0 at max); on the Anthropic side, pair Fable 5.2 with claude-opus-5-5 ($4 / $20 official API price) or claude-sonnet-5-5 ($2 / $10). All of these models are callable on QCode with the same API key.

Sources

OpenAI official docs: GPT-6.2 Astra model page (developers.openai.com/api/docs/models/gpt-6.2-astra), GPT-6.2 Sol model page, OpenAI API pricing page, and GPT-6.2 system card and release notes; Anthropic official docs: Claude Fable 5.2 model overview (platform.claude.com/docs/en/models/fable-5-2/overview), Anthropic API pricing page (platform.claude.com/docs/en/about-claude/pricing), and Claude Code changelog 2.1.321; Artificial Analysis 2026-11-17 leaderboard; QCode: qcode.cc/models and docs.qcode.cc endpoint documentation (all verified on 2026-11-18).

Use GPT-6.2 Astra and Claude Fable 5.2 on One QCode Key

Create a QCode account, billed per token with live per-model rates on /models; run Claude Code and Codex from one shared balance.

Not sure where to start?

Try it for a week

Small Pack: $4 for $30 of quota, valid 7 days

Use it every day

Starter: $9.00/month with $5 of quota every day — just $0.060 per $1 of quota, vs about $0.113 for the Medium Pack